Skip to content

v0.8.3: sim cli, new tools, credential groups, splunk, finder-like files refactor, security hardening - #6765

Open
waleedlatif1 wants to merge 36 commits into
mainfrom
staging
Open

v0.8.3: sim cli, new tools, credential groups, splunk, finder-like files refactor, security hardening#6765
waleedlatif1 wants to merge 36 commits into
mainfrom
staging

Conversation

@waleedlatif1

@waleedlatif1 waleedlatif1 commented Aug 16, 2026

Copy link
Copy Markdown
Collaborator

icecrasher321 and others added 30 commits August 14, 2026 20:10
@icecrasher321 Sim Pi Agent
feat(library): Automation Anywhere Alternative: AI Agents vs. RPA for…
a9688d0 
… Real Reasoning (#6728)

Co-authored-by: Sim Pi Agent <pi@sim.ai>
@waleedlatif1
improvement(credential-groups): align settings surface with the share…
0ce8ded 
…d page patterns (#6727)

* improvement(credential-groups): align settings surface with the shared page patterns

- drop the row "..." menu; a row opening a detail page carries the chevron
  only, and Delete moves to the detail header behind a confirm modal
- replace the hand-rolled Save chip with saveDiscardActions, and wire
  useSettingsUnsavedGuard so detail edits survive tab switches
- fix swapped staleTime constants: the list carried Infinity, which combined
  with the app-wide retryOnMount:false to cache one transient failure until
  a full page reload
- evict the detail query on delete, and keep the bots prop referentially
  stable so a refetch cannot drop a queued Slack authorization message
- reset the detail tab param on open and close so a stale link cannot open
  the next group on the previous group's tab
- match peer rows (iconFilled + --text-icon), drop a bespoke max-w and a
  duplicated gap-7, align no-results copy and the Slack modal field gutter

* improvement(credential-groups): hold first paint for a deep-linked group

Matches the data-drains list: a deep link whose id is still resolving no
longer flashes the list chrome before jumping to the detail. Keys the detail
by group id so lifted draft state can never carry across groups.

* fix(credential-groups): await the refetch before clearing the edit buffer

The update mutation fired its invalidations without returning them, so
mutateAsync resolved before the refetch landed. Callers that clear their
draft on success then fell back onto the pre-save cache and flashed the old
name and description until the refetch completed — or kept showing them if
it failed.
@TheodoreSpeaks
feat(credentials): add v2 credential lifecycle APIs (#6664)
6006870 
* feat(credentials): add v2 OAuth connection APIs

* fix(credentials): preserve active OAuth connection links

* fix(credentials): bind OAuth links to connection intent

* feat(credentials): complete v2 credential lifecycle

* fix(credentials): make disconnect idempotent

* fix(credentials): stabilize oauth draft retries

* fix(credentials): bind oauth callbacks to drafts

* fix(credentials): fail closed on oauth completion

* fix(credentials): bind shopify completion to oauth state

* fix(credentials): align custom oauth reconnects

* fix(credentials): centralize application authorization

* fix(credentials): keep OAuth draft intent immutable

* fix(credentials): allow renamed reconnect targets

* fix(credentials): close OAuth draft edge cases

* fix(credentials): fail closed without breaking auth

* fix(credentials): preserve migrated route behavior

* feat(credentials): add provider search

* fix(credentials): prevent stale secrets and drafts
@TheodoreSpeaks
feat(credential-groups): complete managed account enrollment (#6729)
b198080 
* fix(credential-groups): show reconnect after authorization

* feat(slack): include managed user auth in custom bots

* feat(credential-groups): add enrollment completion page

* fix(credential-groups): align settings order and block color

* fix(credential-groups): delete credentials on access revoke

* fix(credential-groups): use person icon for enrollments

* fix(credential-groups): keep enrollment submit visible

* fix(credential-groups): make enrollment connections optional

* fix(credential-groups): simplify people actions

* fix(credential-groups): preserve pagination after deletion

* fix(credential-groups): delete removed enrollments

* fix(credential-groups): hydrate canvas labels
@TheodoreSpeaks
fix(tables): auto-scroll during column drag (#6722)
cfb99e1 
* fix(tables): auto-scroll during column drag

* fix(tables): keep drag targets aligned while scrolling

* fix(tables): preserve column targets across drag surface

* fix(tables): align workflow group drop indicators
@TheodoreSpeaks @waleedlatif1
@icecrasher321 @claude @Sg312
feat(cli): Sim CLI with AWS-style profiles and a platform key exchange (
337a53f 
#6147)

* improvement(api): pull in the v2 external endpoint surface

Cherry-picks improvement/v2-endpoints (98c85677f5) onto the current base.

The v2 surface standardizes one response family across every endpoint:
`{ data }`, `{ data, nextCursor }`, and `{ error: { code, message, details? } }`,
rendered through apps/sim/app/api/v2/lib/response.ts. v1 auth and rate limiting
are reused as-is; the workspace-access and enterprise-audit checks are split into
`resolve*` cores returning structured failures, with thin v1 wrappers that render
the old `{ error }` body so v1 behavior is unchanged.

The branch's own /api/v2/tables/** is dropped. Staging's tables v2 (#6067,
typed predicate grammar + POST /api/v2/tables/[tableId]/query) supersedes it and
lands in the following merge; the two are reconciled onto the shared envelope
separately.

Conflict resolutions:
- v1/middleware.ts: keeps resolveWorkspaceRequestActor alongside the new
  resolveWorkspaceAccess/resolveWorkspaceScope split
- v1/audit-logs/auth.ts: keeps the newer targetOrganizationId parameter and
  isOrganizationBillingBlocked check inside the structured resolver
- bun.lock: taken from HEAD; the branch's lock churn was unrelated lucide-react
  hoisting

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EsThUPqZXjwuuyRjBbmVkj

* feat(usage): accept X-API-Key on usage-logs list + export

/api/users/me/usage-logs and /export now use checkHybridAuth — the same
auth /api/users/me/usage-limits already accepts — so external monitors
can read summary.bySourceCredits (the source breakdown of usage-limits'
aggregate currentPeriodCost) instead of estimating Copilot spend by
subtraction. Workspace-scoped keys are pinned to their own workspace's
slice of the ledger: the filter defaults to the key's workspace and an
explicit mismatch 403s. Both endpoints documented in openapi-core.json.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CiHhAk2R1NryaS3R8n2yFz

* feat(cli): sim CLI with AWS-style profiles and a platform key exchange

Adds `packages/sim-cli` (`@sim/cli`, bin `sim`) and extends the existing CLI key
handoff so it can mint the credential the public API actually accepts.

## Key exchange

The handoff already existed but only minted *copilot* keys, which do not
authenticate `/api/v1` or `/api/v2` — those want a Sim platform key. The
approval now carries a `scope`:

- `copilot` (the default, so terminals built against the original flow are
  unaffected) mints as before
- `platform` mints a Sim API key: workspace-scoped when the approver is a
  workspace admin, personal otherwise

Scope and workspace are fixed at *approval*, not at poll: the poll is
unauthenticated by necessity, so the browser is the only moment a human is
present to consent and the only place a permission can be checked. The poll
echoes back what was granted rather than what was asked for, so the CLI cannot
file a copilot key under a platform profile and fail later with an opaque 401.

Picking a workspace and scoping a key to it are kept separate. The terminal has
no key yet, so it cannot list workspaces — the browser picker is the only place
that choice can be made, and the pick comes back as the profile's default
whether or not the key is bound to it. Otherwise a non-admin would pick a
workspace by name and then have to go find its id by hand.

Personal-key creation moves into `lib/api-key/orchestration` so the settings
route and the exchange share one issuer.

## CLI

Profiles work like the AWS CLI: `~/.sim/config` for settings (`[profile dev]`),
`~/.sim/credentials` for keys at 0600 (`[dev]`), selected via `--profile` /
`SIM_PROFILE`. Each setting resolves flag → env → file → default, and
`sim whoami` reports the winning source so a surprising value is explainable.
CI can skip login entirely with `SIM_API_KEY` + `SIM_WORKSPACE`.

Commands cover the v2 surface pulled in earlier: workflows, logs, files, and
knowledge, with `--output json` passing the API's own shapes through for `jq`.
`sim tables` is deliberately absent — that surface is still in flux.

## Drift fixes

The v2 routes were authored a month ago and had fallen behind their services:
`checkActorUsageLimits(userId, workspaceId)` → the billing-attribution flow
(which also restores correct payer attribution for workspace keys on KB upload
and search), `processDocumentsWithQueue` gained a required argument, and the
deploy/rollback param objects had stale fields. Caught by a cold type-check —
an incremental run had reported these files clean.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EsThUPqZXjwuuyRjBbmVkj

* feat(billing): dedicated v2 usage endpoints; keep internal usage routes session-only

Replaces the earlier X-API-Key enablement on /api/users/me/usage-logs
with a dedicated public surface, so the internal Billing-settings
endpoints can evolve with the UI while external monitors get a stable
versioned contract:

- GET /api/v2/billing/usage — current-billing-period summary with
  bySourceCredits (the source breakdown external monitors need to watch
  e.g. Copilot consumption without estimating by subtraction), plus
  limitCredits and plan
- GET /api/v2/billing/usage/logs — cursor-paged credit ledger in the v2
  envelope
- workspace-scoped keys are pinned to their own workspace's slice;
  personal keys read the account ledger

The public wire is credits-only: usage-logs rows now carry a hasCost
boolean instead of dollarCost (the Billing UI only needed the >0
signal), and the rateLimit block is removed from the usage-limits
response and docs (deploy-modal tab relabeled accordingly).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CiHhAk2R1NryaS3R8n2yFz

* feat(cli): generate the CLI's v2 API from the route contracts, add tables

The same endpoint was being described in three hand-maintained places: the Zod
contracts the routes validate against, the OpenAPI documents, and the CLI's own
TypeScript interfaces. Two of those are now derived.

## Generation

`scripts/generate-v2-cli-api.ts` reads `apps/sim/lib/api/contracts/v2/**` and
emits `packages/sim-cli/src/generated/v2-api.ts`: request/response types for all
44 operations plus an operation table (method, path, path params) the client
dispatches through, so a route that moves or changes verb moves the CLI with it.

The contracts are the right source because the routes validate against them — a
shape that disagrees with a contract is a shape the server would reject. Zod
4's `z.toJSONSchema()` handles all 110 schema slots; the JSON-Schema-to-TS
emitter is hand-rolled over that known-narrow subset and throws on anything
unrecognized rather than degrading to `any`, since silence is how a generated
client drifts.

`packages/*` must not import `apps/*`, so the generated file is plain type
declarations with no imports and the script does the crossing at build time.

`check:cli-api` fails CI when the file is stale. The generated directory is
excluded from biome: the pre-commit hook runs `check --write`, which would
otherwise reformat generated output and fail that check with an unrelated
message.

## OpenAPI: checked, not generated

The docs specs carry ~1000 hand-written descriptions and ~400 examples that Zod
schemas do not encode, so generating them would trade real documentation for
mechanical accuracy. `check:openapi-drift` reconciles structure instead — every
v2 path and method must exist on both sides — keeping the prose while still
failing on divergence. Both currently agree on all 44 operations.

## Tables

`sim tables list|get|columns|rows|insert|delete-rows`, built on the generated
types. Rows go through the POST query endpoint even unfiltered, since it is the
only shape carrying the predicate. Row columns are discovered at runtime and
unioned across the page, so a sparse row cannot hide a column.

Deletion requires an explicit `--row`/`--filter` selector *and* `--yes`; an
argument-less call would otherwise empty the table. Path params are
percent-encoded — an id containing `/` or `?` would otherwise retarget the
request.

The four existing command groups drop their hand-written interfaces for the
generated ones.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EsThUPqZXjwuuyRjBbmVkj

* fix(cli): make the generated v2 API a fixed point of the formatter

The pre-commit hook rewrote the generated file immediately after it was
committed, so `check:cli-api` then failed in CI reporting contract drift that
had not happened — the only difference was quote style.

The biome.json exclusion added alongside it does not help: lint-staged runs
`biome check --write` on explicit paths, which bypasses `files.includes`. It
implied protection it never provided, so it is removed.

The generator now pipes its output through `biome format --stdin-file-path`
instead, making the emitted file conformant by construction. The hook has
nothing left to change, and the check compares like with like. A formatter
failure throws rather than emitting unformatted output, since falling back
silently would reopen the same loop.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EsThUPqZXjwuuyRjBbmVkj

* fix(cli-auth): wait for the workspace list before allowing approval

The picker fell back to "No workspace (personal key)" while the workspace query
was in flight, and Connect stayed live through that window. A fast click
approved a personal key with no default workspace — when the same click a moment
later would have issued a workspace-scoped key. The fallback read as an answer
rather than a pending state, so the card could promise one outcome and deliver
another.

Connect is now disabled until the list resolves, the trigger shows a loading
label (a placeholder would not show, since the fallback always counts as a
selection), and the explanatory line no longer asserts the personal-key outcome
before it is known.

Failure is treated as degraded rather than fatal: the picker disables but
Connect stays enabled and the copy says a personal key will be issued, so a
transient list failure cannot strand a waiting terminal.

Tests cover the pending, loaded, admin-binding, and error states; the two
loading assertions fail against the previous implementation.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EsThUPqZXjwuuyRjBbmVkj

* fix(cli-auth): name minted keys by timestamp, not date

A second login on the same day failed with `A workspace API key named "CLI
(2026-07-30)" already exists` — after the user had already approved in the
browser, so the whole handoff was wasted and there was no way to complete it
without renaming the existing key.

Key names are unique per owner, so the name has to be unique per login. Now
`CLI (2026-07-30 15:42:07Z)`: second precision, UTC so it is unambiguous in a
shared workspace key list and sorts chronologically.

The comment claiming a same-day collision was desirable (so logins would reuse
one key) was wrong — nothing reuses the key, the mint just fails. A collision at
second precision now means something genuinely unexpected, so it is still
surfaced rather than retried under a suffixed name.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EsThUPqZXjwuuyRjBbmVkj

* feat(docs): validate OpenAPI specs against the Zod contracts in CI

The specs in apps/docs are hand-authored because they carry what Zod
never defines — error envelopes, status codes, prose, examples — so
they can't be generated; check:openapi validates them instead:

- spec integrity: $refs resolve, operationIds unique, 2xx documented,
  no orphaned component schemas
- v2 conventions: every /api/v2 operation documents 401 + 429 and every
  4xx/5xx resolves to the canonical { error: { code, message } } envelope
- contract cross-check: contracts are auto-discovered from
  lib/api/contracts/v2 (each carries its method + path); doc<->contract
  coverage both ways, query/body/response field diffs via z.toJSONSchema
- examples: documented request/response examples must parse with the
  matching contract's actual Zod schemas

First run caught real drift, fixed here: 16 stale orphaned schemas in
the core spec, the v2 billing ops referencing v1-shaped error
components, deploy/rollback examples missing the required nullable
lifecycle keys, CreateTableBody missing folderId, a legacy-grammar
delete-rows example, and four knowledge document ops missing their
required workspaceId query param.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CiHhAk2R1NryaS3R8n2yFz

* fix(docs): recursive field diff in check:openapi + the deep drift it found

A mutation test showed the doc<->contract field diff only compared
top-level properties, so a typo inside the { data } envelope passed.
The diff now descends through matching object properties and array
items (both sides must expose a property set — passthrough contracts
and prose-only docs end the descent instead of false-positive), with
the Zod JSON-schema root doubling as the $defs context.

Deep drift it immediately caught, fixed here: select-column config
(options/multiple) missing from every tables column schema, AddColumnBody
hand-rolling a third column shape (now composed from ColumnInput, with
position/workflowGroupId as the per-op extensions the contracts actually
admit), chunking strategyOptions undocumented, and the deployment
lifecycle fields (activeDeployment/latestDeploymentAttempt) missing from
DeploymentState.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CiHhAk2R1NryaS3R8n2yFz

* fix(security): close the triggerType rate-limit bypass on workflow execute

Caller-supplied triggerType flowed unchecked into preprocessExecution,
whose checkRateLimit default turns OFF for 'manual'/'chat' — so any
API-key caller, and any anonymous public-API caller billed to the
workspace owner, could execute unthrottled by sending
{"triggerType":"manual"} (async runs also skipped the worker-side check
via admissionCompleted). External callers may now only send the
redundant 'api' value; internal JWT callers ('workflow'/'mcp') are
unaffected.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CiHhAk2R1NryaS3R8n2yFz

* refactor(execution): extract enqueue/status/cancel into shared libs

Prepares the v2 execution surface: handleAsyncExecution's queue logic
moves to lib/workflows/executor/enqueue-execution.ts (slot/claim
semantics encoded in a discriminated outcome, not HTTP statuses), the
execution-status read to execution-status.ts, and the order-sensitive
cancel machinery to lib/execution/cancel-workflow-execution.ts. The v1
routes re-render identically — their suites pass unmodified.

Also: preprocessExecution gains rateLimitCounter ('sync'|'async') and
its 429 now carries code RATE_LIMIT_EXCEEDED + retryAfterMs (previously
indistinguishable from the concurrency 429 and Retry-After was
discarded); and the duplicate cancel contract in contracts/logs.ts is
unified on the full 5-value reason enum — its narrower copy made
requestJson throw a client ZodError when cancelling a paused HITL run.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CiHhAk2R1NryaS3R8n2yFz

* feat(execution): callable execution service + structured error classifier

executeWorkflowService composes the same libs the v1 route holds inline
(call-chain guard, execution-id claim, LoggingSession, preprocessing,
deployed-state load + file-field processing, timeout-bound
executeWorkflowCore, output hydration/compaction) for the deployed-state
caller class — the seam the v2 execute route and in-process internal
callers share, making the HTTP endpoint syntactic sugar.

classifyExecutionError stops discarding the block context that
buildBlockExecutionError already attaches at throw sites: failed runs
now yield {message, code, blockId, blockName, blockType} with a stable
append-only code enum (TIMEOUT/CANCELLED/USAGE_LIMIT_EXCEEDED/
INVALID_INPUT/BLOCK_EXECUTION_FAILED/CHILD_WORKFLOW_FAILED/
OUTPUT_TOO_LARGE/EXECUTION_FAILED), so callers route on error class
instead of substring-matching messages — the single place raw errors
are interpreted.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CiHhAk2R1NryaS3R8n2yFz

* feat(api): POST /api/v2/workflows/[id]/execute

Thin route over executeWorkflowService: X-API-Key or anonymous
public-API auth (sync/stream only for anonymous), strict body with
body-flag async (no mode headers on v2), SSE passthrough for stream,
and the execution resource response — executionId always present,
in-band run failures are status:'failed' with the structured
{message, code, blockId, blockName, blockType} error, sync timeout is
status:'failed' + TIMEOUT instead of v1's 408, and a Response block's
payload stays inside output (authors never control response
status/headers on this origin). Async debits the async bucket and the
202 statusUrl points at the v2 executions resource. Adds
CLIENT_CLOSED_REQUEST/SERVICE_UNAVAILABLE to the v2 error codes.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CiHhAk2R1NryaS3R8n2yFz

* feat(api): v2 executions status + cancel with queued backfill

GET /api/v2/workflows/[id]/executions/[executionId] is the single
status URL for sync and async runs: before the async worker writes the
durable log row, status is backfilled from the job queue (deterministic
job id) as 'queued'/'running' — closing v1's 202-to-pickup 404 window —
and failed runs carry the structured error object. POST .../cancel
renders the shared cancellation lib in the v2 envelope with the
tightened 5-value reason enum. Both authenticate via the shared
resolveV2WorkflowAccess (X-API-Key, authz masked as 404,
allowPersonalApiKeys honored).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CiHhAk2R1NryaS3R8n2yFz

* feat(execution): workflow tool + MCP bridge run in-process

workflow_executor (workflow-as-agent-tool) short-circuits in executeTool
through WorkflowBlockHandler — the same invocation boundary canvas child
workflows use — mirroring the deployed_block_executor precedent. The
MCP serve bridge calls executeWorkflowService directly instead of
fetching its own execute endpoint; deployment-version pinning, MCP
response-size rejection, and the actor override become typed options
instead of header sniffing. Both callers drop the double admission slot
and duplicate top-level log row the HTTP hop cost, and failed child
runs now surface the structured error + child executionId so parents
and MCP clients can route on error class and hand providers a
reproducible handle.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CiHhAk2R1NryaS3R8n2yFz

* feat(infra): CORS + CSP coverage for the v2 execute path

/api/v2/workflows/:id/execute gets the same wildcard-origin,
credential-free CORS policy as v1 (the default credentialed policy
would block browser API-key calls and open a cookie CSRF surface) with
X-Sim-Stream-Protocol allowed and no X-Execution-Mode (async is
body-selected on v2), plus the COEP/COOP/CSP header block.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CiHhAk2R1NryaS3R8n2yFz

* feat(ui): deploy modal + copilot advertise the v2 execute surface

All 20 API-tab snippets move to POST /api/v2/workflows/{id}/execute with
the nested {"input": ...} body, async as the "async": true body flag
(X-Execution-Mode gone), status polling against the v2 executions
resource, the third tab renamed Usage and pointed at
/api/v2/billing/usage, and {data} envelope unwraps in the printed
responses. Fixes the latent baseUrl derivation
(endpoint.split('/api/workflows/')) that would have silently built
garbage URLs under a v2 endpoint, and deletes dead code (exampleCommand
across 3 sites, getAsyncExampleTitle). Copilot deploy/manage/serializer
endpoint builders and the api_trigger bestPractices example follow (the
latter also drops its hardcoded staging host).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CiHhAk2R1NryaS3R8n2yFz

* docs(api): document the v2 execution surface

Adds execute, execution status, and cancel to openapi-v2-workflows.json
with the structured ExecutionError schema (append-only code enum + block
attribution) and the ExecutionResource contract, documenting the rules
that differ from v1: modes are body-selected, a failed run is HTTP 200
with status 'failed', an executionId always means data (never the error
envelope), queued status is visible immediately, and Response-block
payloads stay inside output. Registers the three pages in the generated
workflows meta.json and bumps the route-count baseline.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CiHhAk2R1NryaS3R8n2yFz

* feat(api): gate the whole /api/v2 surface behind one flag; UI stays on v1

Every v2 route now runs exactly one check immediately after auth —
v2ApiGateError — and answers 404 when the `v2-api` flag is off, so the
surface is invisible until it is deliberately rolled out. The gate is
keyed on userId only: a workspace/org-keyed check would have to read
membership for a caller-supplied id before authorization runs, and its
404-vs-403 split would leak cohort membership (the trap the per-domain
table gate worked around by running late). The two executions routes
inherit it from the shared access resolver; the tables-specific gate is
removed so no route checks twice.

`tables-v2-api` stays, now gating only the internal predicate-grammar
route /api/table/[tableId]/query — note v2 tables routes move to the
unified flag, so enabling them is a `v2-api` decision now.

Reverts the deploy modal, copilot handlers, and api_trigger example to
the v1 execute endpoint: v1 works unchanged, and the UI must not
advertise a surface most users would get a 404 from.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CiHhAk2R1NryaS3R8n2yFz

* feat(cli): CLI contract for the v2 surface, incl. execution

Adds `packages/sim-cli/src/contract` — the declarative definition of how the
terminal maps onto the API — and folds in the v2 execution endpoints that just
landed on improvement/v2-endpoints.

## The contract

Read it as a diff against what is already derivable, not a listing. Method,
path, path params, field types, enum values, defaults and required-ness all come
from the generated operation table (which comes from the Zod contracts), and the
command name derives from `<resource> [sub-resource] <verb>`. 23 of 47
operations therefore need no entry at all.

The 24 that do carry only what a schema cannot express:
- names, where REST overloads one path — `DELETE /rows` vs `DELETE /rows/[rowId]`
  becomes `batch-delete` vs `delete`, and `DELETE /deploy` becomes `undeploy`
- flags, where a field's type misdescribes its meaning — `workflowIds` is
  `z.string()` that the route splits on commas; no generator can infer that
- columns, which are editorial
- confirm, for the 8 destructive operations

## Execution

`executeWorkflow` / `getWorkflowExecution` / `cancelWorkflowExecution` derive
badly (`/execute` and `/cancel` are verbs the deriver reads as nouns), so all
three are named explicitly: `workflows run`, `workflows executions get|cancel`.

`stream` is marked `omit`: it switches the response to SSE, which the JSON
client would try to parse. Advertising a flag that breaks the response is worse
than not offering it — a `--follow` command that renders the stream is separate
and hand-written, like `files download`.

## Also

- Drops `check:openapi-drift`. The branch landed `check:openapi`, which does the
  same path/method reconciliation plus a recursive field diff and validates doc
  examples against the real Zod schemas — mine was a strict subset.
- Surfaces the new v2 rollout gate in the CLI: it answers 404 for callers
  outside the cohort, indistinguishable from a missing resource, so a 404 now
  carries that as a possibility rather than a diagnosis.
- `executor/utils/errors.ts` widens instead of casting through `unknown`, which
  is both more honest (the value is an Error) and keeps the double-cast ratchet
  at 8.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EsThUPqZXjwuuyRjBbmVkj

* fix(executor): restore child-cost aggregation dropped by the staging merge

Staging's custom-block rewrite deleted `aggregateChildCost` from
workflow-handler.ts, and git merged that file cleanly — but this branch's
workflow-tool-runner.ts, added for the v2 execute migration, still imports it.
A silent semantic conflict: no marker, broken build.

Taking staging's rewrite is correct, so the helper is defined locally in its
one remaining consumer rather than resurrected in the file staging just
rewrote. Same four lines over the still-exported `calculateCostSummary`, so a
failed child workflow keeps billing the hosted-key spend it consumed instead
of reporting $0.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* feat(cli): yaml and text output formats

`--output` now takes table | json | yaml | text, settable per-command, via
SIM_OUTPUT, or persisted per profile as before.

`yaml` joins `json` in rendering the API's raw values rather than the table's
formatted cells, so a duration stays `1500` instead of becoming `"1.5s"` —
switching format changes the encoding, never the data. Line folding is disabled:
valid YAML, but it breaks line-oriented greps and is miserable to read.

`text` is tab-separated with no header and no colour — the shape `cut -f2` and
`while IFS=$'\t' read` expect, so shell plumbing works on a box with no JSON
tool. It uses the rendered cells rather than raw values, since it is a human-ish
format for pipelines rather than something to parse. An absent value collapses
to an empty field instead of the table's em-dash: `cut` returning a literal `—`
would read as a value to every downstream emptiness test.

A bad `--output` is now an error (commander `.choices`) rather than a silent
fall back to `table`. The environment variable and the config file stay tolerant
— those are ambient and set once, so a bad value should not break every command,
but a flag just typed should not be quietly disregarded.

Uses js-yaml 4.3.0, already a direct dependency of apps/sim, rather than adding
a second YAML library to the monorepo.

Also drops a stale README reference to check:openapi-drift, which the v2-endpoints
merge superseded with the deeper check:openapi.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EsThUPqZXjwuuyRjBbmVkj

* refactor(tables): make lib/table/orchestration the single implementation (#6134)

* refactor(orchestration): move the shared error contract out of lib/workflows

OrchestrationErrorCode and statusForOrchestrationError are the contract every
lib/[resource]/orchestration module returns against, but they lived inside the
workflows module, so resource-neutral code (lib/folders) already had to import
from a workflow path. Moved to lib/core/orchestration/types.

Adds a 'locked' class mapping to 423. Both tables and workflows have a lock
that forbids a mutation, and each caller was translating that to a status
itself.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* refactor(tables): make lib/table/orchestration the single implementation

Column update was implemented four times — the UI route, v1, v2, and the
copilot table tool — each calling the same column services but owning its own
guards, error mapping, and audit. The copies had drifted, and the drift was the
bug: v2 was missing both guards, only the copilot copy minted stable option
ids, and only v1/v2 audited.

performUpdateTableColumn, performDeleteTable, and performDeleteTableRow now own
that logic; all ten call sites reduce to auth, parse, call, render. The guards
are asserted once in lib/table/orchestration rather than four times against
four routes.

Behavior this consolidates, previously true on only some paths:

- The typeChanging guard. updateColumnType early-returns on an unchanged type
  and drops any options sent with it, so restating the current type alongside
  new options silently discarded them. v2 had no guard at all and, since its
  contract shares v1's body schema, accepted options and ignored them.
- The select-unique guard. Each write is its own locked transaction, so a
  rename or type change paired with a constraint write that is going to fail
  commits first and then throws, half-applying the schema change.
- Stable select-option ids. Cells reference the option id, so an edit that
  re-sends an option by name has to reuse it or every cell holding it is
  orphaned. Only the copilot path did this; normalizeSelectOptionsInput moves to
  lib/table/select-options and now covers every caller. It preserves a supplied
  id, so it is a no-op for the fully-formed options the HTTP contracts accept.
- required forwarded into the type and options writes, so a conversion
  validates against the constraint the same request is setting.
- An audit on every successful update. The UI route and the copilot tool
  emitted none.
- Single-row delete through the row service. v2 did a raw db.delete, skipping
  assertRowDelete and deleteOrderedRow, so a delete-locked table returned 200
  and the row-count bookkeeping never ran.
- The delete actor handed to deleteTable, which audits only when a row was
  actually archived. v1 and v2 omitted it and audited themselves outside that
  check, emitting TABLE_DELETED for a no-op delete of an archived table.

Failure classes come back as OrchestrationErrorCode; v2 renders them through a
new v2ErrorForOrchestration, mirroring statusForOrchestrationError on the v1
and UI surfaces, so a given failure maps to the same status everywhere.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* test(tables): bind the column-update tests to the orchestration function

The base's route tests assert which column service each payload reaches — the
behavior that now lives in performUpdateTableColumn. They mocked the `@/lib/table`
barrel; the orchestration module imports the service directly, so they mock that
too and keep asserting the same thing through the extracted implementation.

The orchestration tests move onto the base's semantics: writes address the
stable column id, a rename rides inside the write it accompanies rather than
running first, and the currency guards replace the non-select options guard the
service now owns.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* chore(copilot): drop the column-type import the delegation made dead

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* refactor(tables): move the audit log out of the table service

`lib/table/service.ts` wrote its own audit rows, so whether an operation was
audited depended on which function a caller reached for rather than on a user
having performed it. That is what let v1 and v2 audit a no-op delete, and what
made `deleteTable`'s optional `actingUserId` double as an audit opt-out flag.

Worse, most sites fell back to `actingUserId ?? createdBy`, so an unattributed
call was logged against the table's *creator*. The copilot `mv` path passed no
actor at all: renaming someone else's table recorded them as the renamer.

Audit now lives in the orchestration functions — performDeleteTable,
performRenameTable, performMoveTableToFolder, performUpdateTableLocks — and
the services just write. Internal callers (folder cascade, import rollback)
keep calling the service and are silent by construction rather than by
remembering to omit an argument.

Two services now return what the audit needs: `deleteTable` reports whether it
actually archived a row, so a repeat delete logs nothing; `updateTableLocks`
returns the before/after locks, since only the locked write can observe the
transition its description names.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* fix(tables): restore audit provenance and conflict status in orchestration

Moving the audits into the orchestration functions dropped three things the
routes had been carrying, and added one the orchestration now owns twice.

- The v1 and v2 column-update routes passed `request` to `recordAudit`, so
  their audit rows recorded the caller's IP and user-agent. The orchestration
  function had no way to receive it. Every table orchestration function now
  takes an optional `OrchestrationRequestContext` and every HTTP route
  forwards it; the copilot and VFS callers, which have no request, omit it.
- `classifyTableMutation` matched `TableConflictError` on "already exists"
  appearing in the message and reported it as `validation`, turning the UI
  route's 409 on a duplicate table rename into a 400. It now matches the type,
  the way `performRestoreTable` already did.
- `captureServerEvent` ran on every delete while the audit was gated on a row
  actually being archived, so a repeat delete of an archived table still
  reported `table_deleted`. Both now hang off the same evidence.
- The copilot delete path kept its own `captureServerEvent` from when the
  service did not emit one, double-counting every copilot table delete.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YGzbVDZpe2dEALbu2BUU8a

* fix(tables): say which type a no-op column update restated

A copilot `update_column` payload whose only content was the column's current
type used to return success with the live schema, while the v1, v2, and UI
routes rejected the same payload with "No updates specified". Delegating to
`performUpdateTableColumn` unified them onto the routes' rejection — correct,
but the message tells the caller its request was empty when it named a type.

The orchestration function now reports the same thing `updateColumnType` reports
when it loses this race concurrently: the column is already that type, re-issue
without the type change. An empty payload still reads "No updates specified".

Drops the copilot's `outcome.table ?? tableForUpdate` fallback with it — the
comment described the no-op that can no longer reach that line, and a success
always carries a table.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YGzbVDZpe2dEALbu2BUU8a

* refactor(tables): classify failures by type instead of by message text

The table module decided HTTP statuses by searching error messages for
phrases. `VALIDATION_MESSAGE_FRAGMENTS` and `ROW_WRITE_ERROR_PATTERNS` held 32
substrings between them, and fifteen more lists were inlined in routes — 83
matchers over 17 files, each its own copy of the guesswork and already drifted
apart. It made message wording load-bearing: `TableRowLimitError`'s own doc
comment noted that its text had to contain "row limit" for a route to answer
400, and adding "already exists" to a rename message silently demoted a 409 to
a 400 (the bug fixed one commit ago, by adding another special case).

Services now throw `OrchestrationError`, which carries the transport-neutral
`OrchestrationErrorCode` the layers above already speak. Classification is one
`instanceof` in `orchestrationErrorResponse` (UI + v1) and
`v2CaughtOrchestrationError` (v2). Every pattern list is gone. Wording is free
to change; an unclassified error still becomes a generic 500, which is what an
unexpected fault should be.

`asOrchestrationError` walks the `cause` chain rather than testing the caught
value directly: drizzle wraps a throw raised inside a transaction callback in a
`DrizzleQueryError` whose own message is the failed SQL, so a bare `instanceof`
would drop every failure raised inside `withLockedTable`. That is the same
reason `rootErrorMessage` had to dig for a root cause before.

Three throws stay bare `Error` deliberately — `Table ID mismatch`, `Workspace
ID mismatch`, and `Failed to build upsert conflict predicate` are internal
invariants no consumer classified, and they keep falling through to a 500.
`Insufficient capacity` was in the pattern list with no producer anywhere in
the codebase.

Status changes, all deliberate:

- `'forbidden'` joins the code union so the table-row-limit ceiling keeps its
  403; without it this refactor would have flattened it to 400.
- import-async's table-limit rejection: 400 -> 403, matching the two other
  create routes it had drifted from.
- Renaming a table to an invalid name: 500 -> 400. `validateTableName`
  messages don't contain "Invalid", so no matcher ever caught them.
- Restoring a table that isn't archived, or into an archived workspace:
  500 -> 400.
- A duplicate *column* name stays `validation`/400 rather than becoming a 409
  like a duplicate table name. Both v1 and the orchestration have always
  answered 400 for it; changing a published status is not this refactor's job.

The twelve tests that changed were asserting the substring mechanism itself,
constructing plain `Error`s with magic strings. They now assert the real
contract, plus new cases pinning that identical wording carrying no
classification stays internal and keeps its message off the wire.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YGzbVDZpe2dEALbu2BUU8a

---------

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

* refactor(cli): output format is a profile setting, not a flag

Drops `-o, --output`. Format is set once per profile with
`sim configure --set-output <format>`, or overridden ambiently with SIM_OUTPUT
for a one-off (`SIM_OUTPUT=json sim logs list | jq`) and for CI, which already
runs file-less on env alone.

Both remaining sources are ambient — set once, then read by every later command
— so an unrecognized value falls back to `table` rather than breaking the CLI.
There is no longer a strict tier, because there is no longer anything typed
per-invocation to be strict about.

Frees `-o` for `sim files download -o <path>`, which previously had to share the
short flag with a global that meant something else entirely.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EsThUPqZXjwuuyRjBbmVkj

* feat(cli): runtime that builds every command from the contract

Turns the CLI contract into working commands. 43 leaves across 7 groups, up
from the 6 hand-written ones — every v2 operation the contract does not hide is
now reachable, including `sim tables upsert`, `sim workflows run`, and the whole
tables surface.

## What the generator now emits

`V2_OPERATIONS` carries a field→slot map per operation: each query/body field's
kind, whether it is required, its enum values, and its server-side default.
Types alone could not drive this — the runtime has to *iterate* fields to build
flags, and everything from argv arrives as a string, so it needs the kind to
turn "50" into 50 and '{"a":1}' into an object.

It also lifts each operation's one-line `summary` from the OpenAPI specs. The
contracts carry validation, not prose, so `--help` had been showing raw URLs;
the specs already hold a written summary per operation and `check:openapi`
guarantees one exists, so this reuses documentation rather than inventing a
second place to describe the same endpoint.

## The runtime

`derive.ts` names a command `<resource> [sub-resource] <verb>` from the route,
covering 41 of 47. `request.ts` assembles the call: path params from positional
args, `workspaceId` injected from the profile into whichever slot declares it,
everything else coerced and validated locally — so a bad enum, malformed JSON,
missing required flag, or absent workspace fails before any network call.
`build.ts` constructs the commander tree, auto-pages cursor lists up to
`--limit` (0 for everything), and renders through the contract's columns or, for
runtime-shaped rows, keys unioned across the page.

Fixed while wiring: `new Command('upsert <tableId>')` makes the *whole string*
the command name, so `sim tables upsert` never matched and fell through to the
group's help. Arguments have to be declared with `.argument()`.

## What stays hand-written

Two leaves, each for a reason generation cannot satisfy in principle:
`files download` streams binary rather than the JSON envelope, and
`tables rows list` discovers columns from user-defined row data nested under
`data`. They attach onto the generated groups, so `sim files --help` lists them
alongside the rest. The five previous command files are deleted.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EsThUPqZXjwuuyRjBbmVkj

* fix(cli): review round 1 — flag lookup, terminal controls, download safety

## CLI flags silently dropped (Cursor, High)

Commander camelCases every multi-word flag, so `--min-duration-ms` is stored as
`minDurationMs`. `buildRequest` looked flags up by their own kebab name, found
nothing, and dropped the field — no error, it just never reached the API. That
was every multi-word flag on every generated command.

The unit tests passed because they fed flag values already keyed by flag name,
which is not what commander produces — they validated a fiction. Added
`build.test.ts`, which parses real argv through the built commands; three of its
assertions fail against the previous code. The old tests now use camelCase keys
with a comment saying why.

## Terminal control sequences (Greptile, P1 security)

`stripAnsi` matched only SGR (`ESC [ … m`), so a knowledge document, table cell,
or workflow name could carry OSC, non-SGR CSI, or `ESC c` through to an
interactive terminal — setting the window title, moving the cursor to overwrite
what was already printed, or resetting the terminal. Replaced with a `sanitize`
covering OSC (BEL- and ST-terminated), CSI, any ESC + printable, and the bare
C0/C1 range, keeping tab and newline. Applied where API values become display
text, so the colour the CLI adds afterwards still works.

## Downloads (Greptile, P1 ×2)

`createWriteStream` truncated silently, and the destination name usually comes
from the server's content-disposition rather than anything the caller typed —
so a download could irreversibly replace an unrelated local file. Now opens `wx`
and fails with a message naming `--force`, which was added for the deliberate
overwrite.

The stream's error listener was attached after the read loop finished, so an
EEXIST/EACCES/ENOSPC during writing was an unhandled 'error' event that took
down the process. It is now registered before the first write and raced against
the pump.

## Personal-key caption (Cursor, Low)

With "No workspace (personal key)" picked, the caption still promised a default
workspace the approval does not send. It now distinguishes no-pick from
picked-but-not-admin.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EsThUPqZXjwuuyRjBbmVkj

* fix(cli): review round 2 — body-cursor paging, timestamp sanitization

## `tables rows query` printed nothing (Cursor, High)

`isCursorList` only looked for `cursor` on the query slot, but `queryRows` is a
POST whose whole filter — cursor included — is in the body. It therefore took
the single-request path, which handed an array of rows to `printRecord` and
printed an empty record, and it never auto-paged past the first page.

Replaced with `cursorSlot`, which checks both slots and tells the pager where to
put the cursor back. Added a defensive branch so an array reaching the
single-resource path renders as a list with inferred columns rather than
silently printing nothing.

## Invalid timestamps bypassed sanitization (Greptile, P1 security)

`timestamp()` echoes an unparseable value verbatim, and that value is still
server-supplied — so the branch was a way past every other formatter for the
control sequences round 1 closed. Now sanitized on that path too. Audited the
remaining formatters: no other path returns a server value unsanitized.

Both fixes have tests that fail against the previous code.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EsThUPqZXjwuuyRjBbmVkj

* fix(cli): review round 3 — poll retry, download flush errors

Both findings are flaws in round 1's fixes rather than in the original code.

## A redeemable login was thrown away (Cursor, High)

`pollForKey` treated every non-429 status as terminal. But the poll route
releases its mint reservation on any mint failure — its own comment says "a
later poll can retry" — so a transient 5xx or a same-second name conflict ended
the login after the user had already approved in the browser, forcing a full
restart for something the server had deliberately left recoverable.

Retryable is now 409, 429, and 5xx. Everything else stays terminal: 400 means a
malformed request id or verifier and 401/403/404 mean the server is refusing on
purpose, so retrying those would just spin to the 15-minute timeout.

## A failed download reported success (Greptile, P1)

`file.end(resolve)` passes the flush error to the callback as its argument, so
the pump fulfilled *with* the error and the command printed "Saved" for a
truncated file. Confirmed against node directly — `end`'s callback receives the
errno. It now rejects on that argument, which is the path an ENOSPC actually
takes, since the bytes may not reach disk until the final flush.

Adds `device-flow.test.ts` (11 tests: the retry matrix, transport failure,
terminal refusals, and that the poll secret never enters the browser URL) and
`hand-written.test.ts` covering the download's overwrite guard and flush
failure. The two retry tests fail against the previous code; the flush test
needs `/dev/full` and so runs in CI rather than on macOS.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EsThUPqZXjwuuyRjBbmVkj

* fix(cli): review round 4 — repeated flags encode per field kind

`coerce` comma-joined every `list` flag, but that is only correct for the three
fields whose wire type is a `string` the route splits (`workflowIds`,
`folderIds`, `triggers`). The others genuinely want an array:

- `rowIds` and `selectedOutputs` are `array`, so joining sent a string where the
  schema expects a list — `sim tables rows batch-delete --row a b` failed
  validation, and so did a single `--row a`
- `knowledgeBaseIds` is a string-or-array union whose array branch is the right
  one; joining made `kb_1,kb_2` a single bogus id, so multi-`--kb` search
  silently searched nothing

`list` now means only "accept the flag more than once" — the encoding follows
the field's kind, which the generator already records. The two questions were
conflated under one contract field and the `FlagSpec` doc now says so.

Four tests, three of which fail against the previous code.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EsThUPqZXjwuuyRjBbmVkj

* fix(cli): review round 5 — header sanitization, auth ordering, stale suggestion

## Table headers stayed executable (Greptile, P1 security)

Round 1 sanitized cell *values* but not the column *names*, and a table's
columns are user-defined — so the same control sequences were still executable
one row higher, in the header. Sanitizing is now done inside `renderTable`
rather than at each call site, so a future column source cannot reopen it, with
the two key-derived column builders covered as well.

## Fresh install was told the wrong first step (Cursor, Low)

Generated commands read `profile.workspaceId` directly, bypassing
`requireWorkspace()` — which checks the key first precisely so a new user is
told to log in rather than to set a workspace they cannot use yet. That ordering
was fixed for the hand-written commands earlier and reintroduced by the runtime.
`sim tables list` on an empty profile now says "Not logged in" again.

## A stale suggestion shadowed the fallback (Cursor, Medium)

The picker took `selected ?? suggestedWorkspaceId ?? lastActiveWorkspaceId`. The
suggestion comes from a profile the CLI wrote earlier, so it can name a
workspace the user has since left — and merely being truthy, it blocked the
last-active fallback and left the card on "no workspace" with a perfectly good
one available. It now counts only when it resolves against the loaded list.

Two of the three have tests that fail against the previous code; the third is
verified end-to-end (`sim tables list` on an empty profile).

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EsThUPqZXjwuuyRjBbmVkj

* feat(api): add v2 endpoints for MCP servers, skills, custom tools, folders, and credentials (#6150)

* feat(api): add v2 endpoints for MCP servers, skills, custom tools, folders, and credentials

* fix(api): correct credential role, skill permission bar, MCP url identity, and custom-tool conflict mapping

* fix(api): align credential mutation gating, provider-outage status, and unique-violation conflicts

* fix(api): close unique-violation, revival, orphan-write, and env-rename gaps

* fix(api): treat every provider-outage code as unavailable on create and update

* fix(credentials): use the shared outage predicate on the session update path

* fix(contracts): anchor the predicate double-cast annotation to the cast

`check:api-validation:strict` counted 9 unannotated double-casts against a
baseline of 8, failing CI. The predicate leaf schema was annotated, but the
annotation sat above the declaration while the checker anchors on the line
carrying the cast — five lines below, at the close of the object literal. The
scanner walks back at most three lines and stops at the first non-comment one,
so it hit `value: z.unknown().optional(),` and never saw the reason.

Splitting the object schema from the cast puts them adjacent, so the existing
reason binds. No behavior change — the cast, the schema, and the reasoning are
unchanged.

Also lowers the rawJsonReads ratchet 6 -> 5 to match the current count, which
had drifted down; leaving it high lets a removed raw read silently come back.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* fix(skills): point the orchestration error contract at its moved module

#6150 branched before #6134, so skill-lifecycle.ts imports
@/lib/workflows/orchestration/types — the module #6134 moved to
@/lib/core/orchestration/types. Git merged a file deletion on one side with a
new file referencing it on the other: no textual conflict, broken build.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* feat(cli): pick up the new v2 domains; discover modules instead of listing them

Merges `v2-api-spec` (#6150 — v2 endpoints for MCP servers, skills, custom
tools, folders, credentials) and the newer `improvement/v2-endpoints`.

## The generator was list-driven, so none of it would have appeared

`DOMAINS` and `SPEC_FILES` were hardcoded. Five new contract modules and a new
`openapi-v2-resources.json` had landed, and the generator would have skipped
every one — silently, with `--check` still passing, because the generated file
matched a generator that never looked. Both are now discovered from disk.

That is the same silent-drop class the review rounds kept surfacing, and it is
the property the whole pipeline rests on: a new v2 domain should reach the CLI
by regenerating, not by remembering to edit a list.

Result: 47 → 72 operations, 13 contract modules, and 25 new commands
(`sim skills list`, `sim mcp-servers get`, `sim folders delete`, …) with no CLI
change beyond the discovery fix. Summaries for the new domains now resolve too,
so their `--help` reads properly instead of falling back to `METHOD /path`.

## Confirmation gates for the new destructive operations

Five new DELETEs arrived ungated. `deleteFolder` is the sharpest — the route
archives the folder *and cascades to its contents* — so its message says so
rather than reading like a single-item removal.

Added a test asserting every DELETE carries a confirmation, with
`undeployWorkflow` the one documented exception (reversible by redeploying). It
fails against this commit's own starting state, so the next domain to arrive
cannot land ungated the way these did.

## One fix outside the CLI

`lib/skills/orchestration/skill-lifecycle.ts`, added by #6150, imports
`OrchestrationErrorCode` from `@/lib/workflows/orchestration/types`, which does
not exist — the type lives in `@/lib/core/orchestration/types`, where every
other consumer reads it. The branch does not type-check without this.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EsThUPqZXjwuuyRjBbmVkj

* fix(cli): render single-key resource envelopes, and column the new domains

`sim mcp-servers create` created the server, exited 0, and printed nothing.
The v2 route answers `{ data: { mcpServer: {...} } }`, and the record renderer
keeps only scalar fields — one key holding an object left it with none. Unwrap
a lone object-valued key before rendering; a payload with siblings (`{ row,
operation }` from upsert) is a real result and is left alone.

The five domains that arrived with the last generation had no contract columns,
so `mcp-servers list` inferred 20 including `hasOauthClientSecret`. Give each a
column set.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0148a65fkPhP4N8tgGPYtRTU

* refactor(knowledge): make lib/knowledge/orchestration the single implementation (#6154)

* refactor(knowledge): make lib/knowledge/orchestration the single implementation

Knowledge base create was implemented four times — the internal route, v1, v2,
and the copilot tool — and the orchestration around the shared write had
drifted. Extract it the same way lib/table/orchestration was: services write,
orchestration decides which writes run, guards them, audits them, and returns a
transport-neutral failure.

Behavior converged, not preserved:

- One chunking default (DEFAULT_CHUNKING_CONFIG). The agent defaulted minSize to
  1 against the API's 100, so identical input produced differently-chunked
  knowledge bases depending on who created it. The agent path now chunks at 100.
- Every successful mutation is audited inside the orchestration function. The
  copilot tool called recordAudit zero times, so agent-created knowledge bases,
  document uploads, updates and deletes left no audit trail at all.
- Failures classify by class, not by message text. The knowledge service errors
  are OrchestrationError subclasses and storage-quota rejections throw a shared
  StorageLimitExceededError, replacing four separate message greps for
  "already exists" / "does not have permission" / "storage limit".

delete_connector reported the opposite of what happened. It reached the route
through an internal HTTP self-call that sent no query string, so the route's
keep-documents default always applied while the agent told the user the
documents had been removed. The self-call is gone — all four connector
operations run in-process — and the orchestration returns the real counts.

Also:

- OrchestrationErrorCode gains 'payload_too_large' (413 / PAYLOAD_TOO_LARGE).
  Without it, dropping the storage-limit message match would have regressed the
  documented 413 on knowledge base create and document upload to a 500.
- messageForOrchestrationError renders a route's own wording for an unclassified
  fault, so a driver's message no longer reaches the client on a 500.
- v1 and v2 knowledge base update now forward actorUserId, which the service
  requires for a workspace move; both omitted it.
- The connector DELETE route reads deleteDocuments through parseRequest. Its
  contract declared z.boolean(), which would have rejected the string a query
  param actually is.
- Drop the 409 from POST /api/v2/knowledge/{id}/documents in the OpenAPI spec.
  Nothing on the upload path throws a conflict; it was only ever reachable by
  the message match this change removes.

Behavior change worth noting: a v1/v2 PUT carrying only the workspaceId scope
field and no actual updates now returns 400 rather than 200 with the unchanged
knowledge base.

Deliberately deferred: document update remains internal-only. Extracting
performUpdateKnowledgeDocument makes exposing it on v1/v2 a contract and a route
away, but that is a new public surface rather than part of this consolidation.

* fix(knowledge): make connector create atomic and stop flattening failures

Review round 1 on #6154.

- Resolve the billing payer before the connector is committed, not after. A
  malformed attribution header rejected post-commit left a live connector behind
  a 500, and a retry created a duplicate plus duplicate sync work. Manual sync
  resolves before writing its audit for the same reason.
- Let the source-config validator carry its own failure class. Collapsing every
  rejection to `validation` flattened the connector PATCH route's 401 (stale
  stored credential) and 409 (missing workspace context) into a 400.
- Add `unauthorized` to OrchestrationErrorCode. It is the class that 401 was
  already expressing on this route, and the v2 vocabulary already had
  UNAUTHORIZED; only the shared union was missing it.
- Report a knowledge base that exists but failed to archive as failed, with the
  reason, rather than as not found. The copilot delete loop folded every
  non-not-found failure into `notFound`, telling the user it was never there.
- Route copilot failures through the same message helper the HTTP surfaces use,
  so an unclassified fault's raw text (a driver's failed SQL) no longer reaches
  the agent verbatim while the UI and public APIs get the generic wording.

* fix(cli): stop dropping nested fields, and emit exports as documents

`sim workflows export <id>` printed `version` and `exportedAt` and nothing
else. The record builder kept only scalar fields, so `workflow` and `state` —
the entire export — were discarded with nothing to say they had been. Same for
`workflows get`, which silently dropped `variables` and `inputs`.

Record views now render every field. Nested values serialize to one line and
are cut at 160 chars: visibly partial beats silently absent, and json/yaml
output still prints them whole.

Export is a document, not a record — it exists to be redirected to a file and
fed back to `import`, and table/text flatten and truncate, so neither can
round-trip it. `document: true` in the contract makes those formats fall back
to JSON; yaml is honoured because it round-trips.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0148a65fkPhP4N8tgGPYtRTU

* feat(cli): JSON flags accept @file and @- alongside inline JSON

A workflow export is hundreds of lines, and `--workflow` only took it inline.
The shell makes that miserable: unquoted `$(cat wf.json)` word-splits into
broken JSON, and nothing in the help said passing a file was an option.

Every JSON flag now reads `@path`, or `@-` for stdin, so the round trip is
`sim workflows export <id> > wf.json` then `import --workflow @wf.json` — or
one pipe. `@` cannot collide with a real value because JSON only ever starts
with `{ [ " -`, a digit, or t/f/n.

Stdin drains with a readSync loop rather than readFileSync(0): a pipe is opened
non-blocking, so the single-read form returned EAGAIN and died with a raw stack
trace exactly when the upstream process had not written yet.

Parse failures that look like a filename now say so — naming @path, or the file
itself when the bare value turns out to exist.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0148a65fkPhP4N8tgGPYtRTU

* feat(api): expand the public v2 files surface (#6160)

* feat(api): expand the public v2 files surface

Adds folder support, rename/restore, move, bulk archive, share, and content
replace to /api/v2/files, so managing files by API no longer stops at
upload + download + archive-one.

Routes are thin: auth -> parse -> perform* -> serialize. Share and content
replace get their orchestration extracted first so the session routes and
the public ones cannot diverge on the effective-authType resolution, the
EE public-sharing gate, or the storage-quota classification.

Presigned upload stays session-only: presign does an advisory quota check
and the real debit happens in the separate register step, so a caller that
never registers leaves unaccounted bytes with no reaper. The buffered
multipart path debits inside uploadWorkspaceFile's own transaction.

* fix(files): classify folder and content failures instead of 500ing them

Bugbot round 1. The v2 routes map errorCode straight to a status, so every
manager failure that arrived unclassified became a 500 for what is really a
caller-fixable 400 or 404.

- Folder manager throws OrchestrationError: missing target/folder -> not_found,
  reparent cycle / self-parent / restore-into-archived-workspace -> validation.
- File manager does the same for the in-transaction 'File not found' paths that
  the earlier pass missed.
- updateWorkspaceFileContent's outer catch re-wrapped everything in a bare
  Error, which stripped the class off StorageLimitExceededError and the new
  not_found alike. It now rethrows a classified failure untouched and attaches
  cause to the generic wrap, so asOrchestrationError can still walk the chain.
- Every remaining perform* gained the asOrchestrationError branch.
- renameWorkspaceFile returned the pre-update read, so the v2 PATCH reported a
  stale updatedAt; it now returns the timestamp it actually wrote.

Docs: upload auto-suffixes a duplicate name rather than rejecting it, matching
the in-app uploader. The description claimed 409 and was simply wrong.

* fix(files): surface a failed upload read-back as the real error

getWorkspaceFile swallows a query failure and returns null unless throwOnError
is set, so a transient blip on the post-upload read reported as 'file could not
be read back'. Distinguish the two: a real null after a just-committed write is
an invariant break, a query failure is itself.

* revert(api): drop the dedicated v2 file-folder routes

File folders already live in the shared folder table as resourceType 'file'
(#6045 cut them over, #6051 dropped workspace_file_folders), and the remaining
file-specific folder machinery is being folded into the generic folder engine.
Publishing /api/v2/files/folders/** would pin that transitional split into a
public contract we'd then have to keep or break.

Files stay folder-aware — folderId/folderPath on the projection, folderId on
upload, and the move route — because a folder id is a folder.id and survives
the unification untouched. Folder management belongs on /api/v2/folders once
that surface serves resourceType 'file'; until then there is no v2 way to
enumerate file folders, which is the deliberate gap.

The orchestration classification fixes stay: the internal routes and the
copilot file-folder tools still call those perform* functions.

* fix(files): classify upload failures instead of matching their wording

Bugbot round 2. uploadWorkspaceFile had the same outer-catch rewrap that
updateWorkspaceFileContent did, so a blown storage quota reached the route as a
bare Error and the v2 handler recovered the status by substring-matching the
message. Any rewording silently demoted a 413 to a 500.

- uploadWorkspaceFile rethrows a classified failure untouched and attaches cause
  to the generic wrap.
- FileConflictError is now an OrchestrationError('conflict'), so a duplicate name
  classifies like every other conflict. Its 'FILE_EXISTS' discriminator had no
  readers and is gone; the instanceof checks elsewhere still hold.
- The v2 upload handler uses v2CaughtOrchestrationError, dropping all three
  string matches.

Also documents that bulk-archive is best-effort: unknown or already-archived ids
are skipped rather than failing the call, and deletedItems is what actually
happened. That asymmetry with the single-id DELETE was undocumented.

* feat(cli): wire the expanded v2 files surface

Regeneration picked up seven new operations (72 → 79), every one of which
derived badly. `/files/move` and `/files/bulk-archive` put a verb where the
deriver expects a sub-resource, so each became a group holding a lone `create`;
`GET /files/[id]/share` fetches one share a…
@TheodoreSpeaks
fix(credentials): conceal inaccessible credential reads (#6730)
c269e88
@waleedlatif1
feat(resources): multiselect on tables and knowledge, spring-loaded f…
0cd87ba 
…olders (#6721)

* feat(resources): multiselect on tables and knowledge, spring-loaded folders

Tables and Knowledge lists get the checkbox multiselect Files already had —
selection, shift-click ranges, select-all, and a shared bulk action bar for
move and delete.

Dragging a resource onto a folder row and resting there now opens that folder,
so nested filing is one gesture (macOS Finder spring-loading). Works on Files,
Tables, and Knowledge.

Selection, the action bar, the drag payload, the drag ghost, and drag teardown
are extracted to shared modules; Files migrates onto them rather than keeping
its own copies.

Bulk move and delete land as single authorized operations that take folders and
resources together, so a mixed selection commits once instead of fanning out.

Fixes two latent UI bugs: the drop-target outline referenced --accent, an
HSL-channel token only valid via hsl(), so it silently rendered as currentColor;
and rows painted hover and selected with the same surface token, making the two
states indistinguishable.

* chore(audits): re-record route ratchet after merging staging

* fix(bulk): reject a move target inside the moving subtree and report contained folders deterministically

* improvement(resources): neutral drop affordance, longer spring delay, and a body drop target

* fix(resources): drop into the open folder on Files, return on an unused spring-open, and guard bulk caps

* fix(drag): end a drag on pointer resume instead of an idle timer

* feat(resources): drag onto breadcrumbs to move back up, and round the drop ring

* fix(resources): tint the list region on drop instead of ringing it

* refactor(folders): share the spring-navigation lifecycle so Files returns too

* fix(breadcrumbs): accept a drag on the open-folder crumb too

* fix(files): keep the view in a spring-opened folder when an OS upload lands there
@waleedlatif1
fix(knowledge): bound chunking separators so one config can't stall p…
f0fd48c 
…rocessing (#6735)

* fix(knowledge): bound chunking separators so one config can't stall processing

`chunkingStrategyOptionsSchema.separators` accepted an arbitrary-length array
of arbitrary-length strings, next to a `pattern` field already capped at 500
chars. `RecursiveChunker` splits the whole document once per separator and
walks the list from the top for every oversized fragment, so a persisted
config with thousands of non-matching separators cost seconds of synchronous
CPU on every later document upload — work neither the processing `Promise.race`
timeout nor the after-the-fact chunk-count cap can interrupt. Measured on a
21.3 MB document: 632 ms at 100 separators, 6.1 s at 1000, 36.8 s at 5000.

- Bound `separators` to 32 entries of at most 100 characters on the write path.
  The largest built-in recipe (markdown) uses 16, so hand-tuned lists still fit.
- Keep the stored/read shape tolerant, so a config written before the bound
  still lists instead of failing response validation.
- Clamp in `RecursiveChunker` too, with a warning, so an already-persisted
  oversized list cannot reach the split loop. An over-long separator is dropped
  rather than truncated: a truncated separator matches where the configured one
  never did, silently re-cutting the document, while dropping it behaves like a
  separator that finds no match. A list left empty falls back to the recipe.
- Walk non-matching separators iteratively instead of recursing, so stack depth
  no longer tracks the separator count. Verified behavior-preserving against the
  previous implementation over 4000 randomized configs — byte-identical output.
- Validate in the create-base modal so the limit surfaces inline.

After the fix the same 21.3 MB document costs ~300 ms at every separator count.

* fix(knowledge): gate separator validation on the recursive strategy

- The separator refines ran for every strategy, but the field only renders for
  `recursive` and only that strategy submits it, so a value left behind by a
  strategy switch could block submit with no visible field to clear. Gated the
  same way the regex-pattern refine already is.
- Use absolute imports in the chunker test, per the repo convention.
@TheodoreSpeaks
fix(docker): prune the app package by manifest name (#6736)
e3b428e
@waleedlatif1
fix(credentials): let workspace admins disconnect a teammate's OAuth …
03ba95a 
…credential (#6737)

* fix(credentials): let workspace admins disconnect a teammate's OAuth credential

Disconnecting an OAuth credential routed through POST /api/auth/oauth/disconnect,
a credential *user* operation scoped to the acting user's own `account` rows. A
workspace or org admin acting on a teammate's connection matched no accounts, so
the call returned `{ credentials: [] }` and the route still answered 200 — the UI
navigated away and the credential was still there. Not a denial, a silent no-op.

Route every type through the workspace-scoped credential delete instead, which
authorizes against credential admin and already resolves workspace and org admins
as derived credential admins for shared credential types.

That path only deleted the `credential` row, so send `oauth` through
`deleteCredentialRecord` — the manager that also tears down a credential's secret
source — and teach it to revoke the backing `account` grant once no credential
references it. Scoped by account id, not owner: the caller is already authorized
against the credential, and the grant belongs to the teammate.

Also make RoleLockTooltip layout-transparent. It wrapped locked controls in an
`inline-flex` div, which let the chip shrink to its label while unwrapped
controls stretched to the member row's fixed role track — so a credential's own
members list rendered Admin at two different widths. A `grid` wrapper stretches
like the unwrapped control, aligning the credential, secrets, and skills member
lists that share the row.

* refactor(credentials): consolidate credential deletion onto one path

Follow-ups from review of the previous commit.

- Collapse the delete use case's remaining `service_account` carve-out. Both
  ternary arms reached the same `deleteConnectionCredential` tail, but the
  carve-out skipped `deleteCredentialRecord`'s Slack custom-bot guard — and
  custom bots are exactly the type it guards, so the single-delete surface
  could orphan a credential group that the batch path refuses to.
- Make `deleteOrphanedOAuthAccount` one conditional statement instead of a
  read then a write. `credential.accountId` is ON DELETE CASCADE, so a
  credential racing the gap would have been reaped by Postgres without
  `clearCredentialRefs` running, stranding its id in workflow state.
- Give oauth its own branch in `deleteCredentialRecord`, matching the shape
  the env types already use, rather than a conditional tail after the return.
- Point `handleReconnectCredential` at the shared helper; it carried its own
  copy of the same orphan-grant rule.
- Invalidate the OAuth connections query on credential delete. The removed
  disconnect hook owned that invalidation, and the detail page reads it. This
  also retires the hand-dispatched `oauth-credentials-updated` event from the
  delete path; the connect/reconnect path still dispatches it for its listener.
- Delete `useDisconnectOAuthService`, now callerless.
- Trim comments to the behavior rather than the bug that motivated it.
@waleedlatif1
fix(media): prevent drawtext filtergraph injection in add_text (#6734)
b96c053 
The add_text FFmpeg operation inlined the caller's caption into a
single-quoted `drawtext=text='...'` filter option. FFmpeg's av_get_token
copies bytes verbatim inside a single-quoted run, so a literal quote in
the caption closed the quote and the remainder was parsed as filtergraph
syntax. An attacker could inject `drawtext=textfile=<path>` (arbitrary
local-file read) or `movie=filename=<url>:f=tty` (read-SSRF), rendering
the target file bytes or HTTP response body into the returned video.

Route the caption out-of-band: write it to a file the operation owns and
reference it via `drawtext=textfile=caption.txt` with `expansion=none`,
so the caption bytes never re-enter the filtergraph parser. The caption
is referenced by a bare relative filename with FFmpeg's working directory
set to the temp dir, because FFmpeg's tokenizer cannot round-trip a
single quote inside a textfile= value — an absolute temp path would break
add_text whenever os.tmpdir() contains a quote (e.g. a Windows profile).
The working directory is passed via execve and never parsed as graph
syntax, so any character in it is safe.
@waleedlatif1
fix(auth): scope SSO account linking to the verified domain and fence…
77f520c 
… plugin provider mutations (#6738)

* fix(auth): fence off plugin-served SSO provider mutation endpoints

The auth catch-all forwarded every non-organization POST to the better-auth
SSO plugin, leaving sso/update-provider and sso/delete-provider reachable
alongside the app-owned sso/register route.

update-provider is gated only on provider ownership and merges the caller's
samlConfig, so a provider owner could set mapping.emailVerified — a field the
register contract deliberately omits and the plugin's identity-boundary guard
does not inspect, so it never trips the linked-account conflict. With
trustEmailVerified enabled, a subsequent assertion carrying an arbitrary
verified email auto-links to that user's account.

Block SSO POST paths by default, allowing only the sso/saml2/ protocol
endpoints the IdP posts to, mirroring the existing organization fence.

* fix(auth): stop trusting IdP email_verified for SSO account linking

Better Auth's link gate is `!isTrustedProvider && !userInfo.emailVerified`,
so trustEmailVerified let a true email_verified claim stand in for the domain
binding. Any principal able to register an SSO provider — an Enterprise org
admin, or any signed-in user when self-hosted — could point it at an IdP they
control, assert an arbitrary victim's address as verified, and auto-link into
that account across tenant boundaries, persisting as an account row.

With it off, linking requires isTrustedProvider, which is domainVerified plus
validateEmailDomain(email, provider.domain) — a provider can only claim
identities inside the domain it proved. That is the model the codebase already
documents for trustProviderByName: false.

The option only ever set emailVerified on the local row; it was never what
made linking work, since Entra omits the claim and SAML ignores it without a
mapping the register contract does not accept.

* docs(auth): note that the SSO fence and trustEmailVerified are layered

* test(auth): drop unnecessary any casts from the auth catch-all tests

createMockRequest already returns a NextRequest and the handler mocks are
untyped vi.fn()s, so every cast in the file was suppressing type checking for
no reason.
@waleedlatif1
refactor(resources): converge Files onto the shared drag hook and bat…
4fc0fb4 
…ch bulk authorization (#6748)

* refactor(resources): converge Files onto the shared drag hook and batch bulk authorization

Files kept a 280-line copy of the foldered-list drag logic because it also
accepts OS file drops. The copies had already drifted, so the external drop
becomes an option on the shared hook and the copy goes away.

- Add `externalDrop` to `useFolderRowDragDrop`: folder rows highlight and spring
  open for an OS file drag exactly as for a move, while the body and breadcrumb
  decline so the page-level upload overlay owns those regions
- Collapse the three drop-active booleans into one `ActiveDropTarget` union, so
  exactly one affordance is armed by construction rather than by hand-clearing
- Keep drop-target writes identity-stable so `dragover` does not re-render the
  list on every event
- Give each list its own drag MIME again, restoring the cross-surface isolation
  `drag-payload.ts` documents
- Let a folder spring open more than once per drag, so a drag can walk back out
  through the breadcrumb and descend again; the guard against re-entering the
  folder already on screen moves to `useSpringNavigation`, the only layer that
  can state it
- Resolve each bulk item against the workspace context the batch already holds,
  and memoize the effective-permission lookup for the batch, replacing two
  invariant queries per item
- Fill the drop target at `--surface-active`: `--surface-4` is the button-base
  token and is lighter than hover in light mode, so the strongest row state read
  the faintest

* fix(resources): dismiss the upload overlay on a folder drop and re-check permission per item

The drag hook stops propagation on a drop it handles, so the page-level handler
that cleared the upload overlay never ran and the chrome stayed up over the
finished upload. Both consuming paths now share one dismissal.

Drop the batch permission memo: each item in a bulk move or delete commits
independently, so reusing one allow verdict let a revocation part-way through a
batch go unseen by the remaining items. The workspace context is still resolved
once per batch, which was the larger saving.
@waleedlatif1
feat(okta): add System Log, MFA, sessions, apps, roles, and group rul…
d45dad7 
…es (#6741)

Expands the Okta block from 18 to 44 operations, covering the System Log, MFA
factors, sessions, applications, administrator roles, and group rules.

Adds shared helpers for the SSWS auth header, Okta error parsing, and the
Link-header `after` cursor, and routes every tool through them so there is one
auth and error path. All eight list operations now return `nextCursor` and
`hasMore`.

Makes the block's param transform authoritative over the serialized inputs: the
executor merges it on top of them, so a key the transform omits keeps the raw
subBlock string. Assigning `undefined` is what actually drops it, which is what
keeps a non-numeric `limit` from reaching Okta verbatim and stops a blank field
in a partial `update_user` from overwriting the stored value with an empty
string.
@waleedlatif1
improvement(tables): make Cmd+F search as you type and clear on close (…
611df8b 
…#6733)

* improvement(tables): make Cmd+F search as you type and clear on close

* fix(tables): reset the find debounce on close and land Enter on a skipped first match

* fix(tables): keep a clicked cell selected when find closes

* fix(tables): strand an in-flight match jump when the search term changes

* fix(tables): strand match jumps on the live query, not the debounced one

* fix(tables): cancel before reveal, and release find's selection on any grid key

* fix(tables): drop find's selection restore, commit the term on Enter

* fix(tables): block find navigation until the results describe the term

* fix(tables): clamp the find step base when a refetch shrinks the match set

* fix(tables): track the find cursor by match identity, not position

* fix(tables): release the find cursor when its match leaves the result set

* fix(tables): skip the reveal when the target match vanishes mid-jump
@waleedlatif1
feat(splunk): add Splunk Enterprise and Cloud integration (#6743)
852906e 
Adds a Splunk block with 12 REST operations: run search (oneshot), create/get/cancel search job, get search results, list/get/dispatch saved searches, list/get fired alerts, list indexes, and list apps. Bearer-token or basic auth, with optional /servicesNS namespace scoping.

Every tool was validated against the Splunk REST reference. Results use search/v2/jobs/{sid}/results because the v1 endpoint is deprecated and disabled from Splunk Enterprise 9.0.1. A half-specified namespace fills the missing node with the documented - wildcard rather than nobody/search, which would have hidden user-private objects. Dispatching endpoints fail loudly instead of reporting success with a null sid, and Create Search Job rejects exec_mode=oneshot since that mode returns results rather than a search ID. The results and control endpoints tolerate an empty body. saved/searches sends the f field filter the reference prescribes for it.
@waleedlatif1
fix(okta): stop partial updates erasing stored profile data (#6751)
cbbcca9 
* fix(okta): stop partial updates erasing stored profile data

Post-merge audit of the Okta integration (follows #6741), verified against the
OpenAPI spec bundled in okta-sdk-golang/.generator.

Two updates could silently destroy data:

- `update_group` targets `PUT /api/v1/groups/{groupId}`, which Okta documents as
  `replaceGroup` — it swaps the profile wholesale. Sending only the two fields
  the tool exposes erased the stored description on every rename, and dropped
  every org-defined custom attribute along with it. The tool now reads the group
  and overlays the supplied fields before replacing, matching the read-modify-
  write `salesforce_update_custom_field` already uses for the same hazard.
- `update_user` gated its profile fields on `!== undefined`, so an empty string
  reached Okta and blanked the stored value. The block strips blanks before they
  get there, but the tool is `user-or-llm` and a model routinely emits `""` for a
  field it has nothing to say about, so the guard belongs on the tool.

Also corrected:

- `forgetDevices` defaults to true at Okta, so the unseeded switch rendered off
  while remembered factors were in fact being cleared.
- Group rules take a plain keyword on `search`, not the SCIM-style expression the
  shared Search field's wand generates, so they get their own field.
- `get_logs` dropped `limit=0`, which the spec documents as valid.
- `get_user` emitted an activation timestamp under `activated`, which the block
  declares as the lifecycle boolean; the timestamp is now `activatedAt`.
- Descriptions that overstated what an endpoint does: `list_users` omits
  DEPROVISIONED users, `delete_user` deactivates before it deletes,
  `delete_group_rule` answers 202, and `excludedGroupIds` is always empty because
  Okta does not support group exclusions.

* fix(okta): forward the abort signal through the group read-modify-write

* test(okta): rename the shared body-builder helper

* fix(okta): key the send-email and search mappings off the operation

* docs(okta): use TSDoc for the new block annotations
@TheodoreSpeaks
fix(credentials): restore reliable OAuth connections (#6753)
623c30f 
* fix(credentials): restore reliable OAuth connections

* fix(credentials): cover OAuth handoff edge cases

* fix(credentials): preserve reauthorization outcomes
@waleedlatif1
fix(deploy): resolve the error-output flag from edges on both sides o…
4f722c6 
…f change detection (#6754)

* fix(deploy): resolve the error-output flag from edges on both sides of change detection

A block can hold `errorEnabled: false` while an error edge still leaves it.
`setBlockErrorEnabled` does not remove existing error edges, both block renderers
draw the port on `errorEnabled || hasErrorConnection`, and the executor never
reads the flag at all — the edge alone decides routing. So the two spellings are
one state, and nothing about the block has functionally changed.

Only the deployed side reconciled them. `materializeDeploymentState` backfills
`errorEnabled: true` for any block with an error edge, while the live normalized
tables are read verbatim. Change detection compared the raw flag, saw `true`
against `false` for a block that had not changed, and no redeploy could clear it:
deploying snapshots the live `false`, which the next read backfills straight back
to `true`. The deploy button sat on "Update" permanently, and the server path
(`checkNeedsRedeployment`, which reads the raw jsonb and skips the backfill)
disagreed with it.

Lift the rule into `@sim/workflow-types` as `collectErrorSourceBlockIds` /
`resolveEffectiveErrorEnabled` so the backfill and the comparison share one
definition, and apply it to both sides of the diff. Compared outside the
structural gate, since the flag can match while the edges disagree.

* chore(deploy): use TSDoc for the error-output comparison comments
@waleedlatif1
fix(tables): re-check the find match at the reveal, not just before p…
0077f6f 
…aging (#6750)

* fix(tables): re-check the match at the reveal, not just before paging

* fix(tables): only release the find cursor from the reveal that owns it
@waleedlatif1
fix(fork): preserve folder structure across a fork edge for files, ta…
76318e4 
…bles, and knowledge bases (#6752)

* fix(fork): carry folder structure across a fork edge for files, tables, and KBs

Only workflow folders were mirrored into the target workspace on fork
create and on sync. Copied files, tables, and knowledge bases were written
with a hardcoded `folderId: null`, so a push or pull flattened them all
into the target root and lost the source's grouping — visible as a fork
sync that drops folder structure when copying files to the parent.

`resolveForkFolderMapping` already did the real work (prune to folders
holding copied content plus ancestors, reuse same-named target folders,
remap parentId), but was pinned to `resourceType: 'workflow'` on both
reads and on the folder-ceiling check. Parameterize it by resource type
and run it per family, threading the resulting map into each copy instead
of nulling. The four folder-bearing families own disjoint trees and folder
ids are globally unique, so the per-family maps merge cleanly for the
`sim:folder/<id>` content rewrite, which previously resolved only for
workflow folders.

Existing forks are healed on their next sync rather than by a migration:
`rehomeFlattenedForkResources` re-homes mapped files/tables/KBs whose
target `folder_id` is still NULL — the exact signature of the old
flattening — so a placement chosen in the target is never overwritten and
the pass converges to a no-op.

`BlobCopyTask.targetFolderId` is optional so tasks queued by an earlier
deploy replay at the root exactly as before.

* refactor(fork): page the re-home lookups and reuse the plan's identity rows

Self-review of the folder-transit change surfaced two scaling problems in
the re-home pass, both of which grow with the size of the fork edge rather
than the size of the sync:

- The resource lookups built `IN (...)` lists straight from the edge's
  mapping rows, so a large fork could hand Postgres a list approaching the
  bind-parameter ceiling and a pathological query plan. Page them at 500,
  matching the paging the rest of the fork copy already uses.
- The pass re-read the whole edge mapping via `getEdgeMappingRows`, which
  the promote plan had already loaded in the same transaction — a second
  full load of identical rows. Expose them on `ForkPromotePlan` and pass
  them in, which also drops a mock from the re-home tests.

Also tally moved rows from `returning()` rather than the planned batch
size, so the log line reports what the `folder_id IS NULL` guard actually
wrote instead of what was attempted.

* fix(fork): drop the sync-time re-home pass, keep folder transit forward-only

Review surfaced three findings and every one of them was in the re-home
pass, none in the forward-looking fix:

- It keyed mapping orientation off `direction`, but the promote route
  resolves the edge from whichever workspace the caller is acting in, so a
  caller in the PARENT pushing to its child is `direction: 'push'` with the
  parent as source. The plan derives this as
  `sourceWorkspaceId === edge.parentWorkspaceId` for exactly that reason.
- Moving a file into a mirrored folder can violate
  `workspace_files_workspace_folder_name_active_unique`, which would abort
  the whole promote transaction and take the workflow sync down with it.
- `folder_id IS NULL` cannot distinguish "flattened by the old copy" from
  "the user moved this to the root", so the pass re-applied on every sync
  and would fight a deliberate placement indefinitely.

The first two are fixable; the third is not without a one-time marker per
edge, which means a migration. A heal that re-applies forever is worse
than no heal, so remove the pass entirely rather than ship it half-right.
Folder structure now transits correctly from this point forward, which is
the actual reported bug; healing already-flattened resources can be a
separate change with a marker to make it run exactly once.

Reverts the `ForkPromotePlan.mappingRows` field with it — it existed only
to feed this pass.
@waleedlatif1
feat(crowdstrike): add alerts, host response, IOC, Spotlight, RTR, an…
4bc89c9 
…d case tools (#6746)

* feat(crowdstrike): add alerts, host response, IOC, Spotlight, RTR, and case tools

CrowdStrike Falcon shipped only three read-only Identity Protection sensor
tools. This adds 20 tools across the response and investigation surface SecOps
teams actually automate against.

Alerts (current Alerts API): query, get details, update status/assignment/
tags/comment/visibility. Hosts: contain, lift containment, hide, unhide. Host
groups: query, get details, add/remove hosts. IOC Management: query, get, create,
update, delete. Spotlight: query vulnerabilities, get vulnerability details. Real
Time Response: init session, execute a read-only command, poll command status,
delete session. Case Management: query cases, get case details.

Every endpoint, request field, and response field is taken from CrowdStrike's
published surface (developer.crowdstrike.com API reference, FalconPy endpoint
definitions, and the swagger-generated gofalcon models). Required API scope is
documented in each tool description.

Deliberately not implemented:
- Detects API: decommissioned 2025-09-30, superseded by Alerts.
- CrowdScore Incidents API and behaviors: decommissioned 2026-03-09 and removed
  from the developer center entirely. Case Management is CrowdStrike's
  replacement, so its two documented read operations are implemented instead.
- Case create/update/merge: the swagger types case `status` and
  `severity_info.level` as bare strings with no enum, so a correct write cannot
  be built without guessing.

CrowdStrike answers 200 with a populated `errors` array for partial failures.
Responses now surface those per-item errors, and an empty result set carrying
errors is reported as a failure rather than silently succeeding.

The route's shared Falcon client, response normalizers, and operation dispatch
move into colocated modules so the handler stays readable at 23 operations.

* fix(crowdstrike): correct the RTR read-tier commands and stop dropping the IOC delete filter

Validation pass over all 23 tools against CrowdStrike's swagger-generated SDKs
(gofalcon falcon/models + falcon/client, FalconPy _endpoint/*.py) turned up four
real defects.

The Execute RTR Command dropdown offered `csrutil` and a bare `reg`. Neither is
a read-tier base command: CrowdStrike's own swagger description for
RTR_ExecuteCommand enumerates cat, cd, clear, env, eventlog, filehash, getsid,
help, history, ipconfig, ls, mount, netstat, ps, and "reg query". `csrutil`
appears nowhere in CrowdStrike's published surface, and `reg` alone is not a
base command — the registry variants are "reg query" (read) and "reg set"/"reg
delete" (Active Responder). Both entries are corrected everywhere they were
repeated: dropdown, tool description, and param description.

Delete Indicators showed a Filter input, declared the param, accepted it in the
contract, and implemented CrowdStrike's documented filter-takes-precedence rule
in the route — but the block never mapped the field into the tool call, so the
filter was silently discarded and a filter-only delete failed validation. The
`filter` case is now mapped alongside the ID list.

A 200 carrying only envelope errors was reported as HTTP 200 with success:false,
which reads as a success to anything inspecting status. Failures now adopt the
per-item error code the envelope supplies, falling back to 502.

Alert updates gain a first-class Remove Tags By Prefix field. The spelling was
previously unresolvable, so it was left to the raw action-parameter escape
hatch; CrowdStrike's swagger settles it as `remove_tags_by_prefix` in both the
PatchEntitiesAlertsV2 and PatchEntitiesAlertsV3 descriptions.

Case Management and Spotlight scopes now name the OAuth scope string
(case-templates:read, spotlight-vulnerabilities:read) alongside the label the
Falcon API client UI shows, so either rendering is findable.

* fix(crowdstrike): stop blank sensor filters reaching Falcon and expose the RTR outputs

The falcon.ts/normalize.ts/operations.ts split routed query_sensors through the
shared buildUrl helper, which skips only undefined. An empty filter or sort
string therefore emitted `?filter=` / `?sort=` where the pre-split route omitted
the param, sending Falcon an empty FQL expression. Reject blank values in the
contract instead, matching the newer operations.

Also surface the ten RTR fields the tools already return but the block never
declared, and broaden the block metadata past the original sensor-only surface.

* fix(crowdstrike): fail query operations on error-only envelopes and guard IOC pagination

Falcon can answer 200 with an errors array and no resources. The detail
operations already treated that as a failure, but the five query branches
returned an empty successful result, so a failed alert query read as a valid
no-match to the calling workflow.

Blank FQL rejection now covers the alert, host-group, indicator, vulnerability,
and case contracts too, not just sensors, and Query Indicators rejects offset
combined with after instead of forwarding a pagination pair CrowdStrike refuses.

* fix(crowdstrike): stop blank inputs reaching Falcon and restore the dropped output docs

The executor merges `tools.config.params` over the raw block inputs, so a key the
mapper omitted kept its raw subBlock value — and an untouched subBlock is stored
as `null`, which the route contract rejects. Query Alerts with an empty Filter,
Update Alerts without every optional field, and Delete Indicators without an
audit comment all 400'd before reaching CrowdStrike. Seed every optional key as
`undefined` so omission is authoritative, which also stops a value left over from
another operation riding along.

Shared output consts in `outputs.ts` were silently dropped from the generated
docs: the generator scans tool source and resolves consts only from `types.ts`,
so `errors`, `affected`, and `pagination` rows vanished from 17 tool pages and
every nested property row with them. Inline the literals.

Against CrowdStrike's own generated SDKs and developer portal:
- add csrutil, ifconfig, users, and the eventlog subcommand forms to the
  read-tier RTR base commands, matching PSFalcon's ValidateSet
- add detection_suppress/detection_unsuppress and cap host actions at the
  documented 100 ids
- cap the IOC search limit at the documented 500, not 2000
- correct the Cases scope to "Cases: Read"; case-templates guards a different
  collection
- type the IOC payload so a blank string cannot clear a stored field on PATCH
- send `MsaRangeSpec` bounds capitalized, as the spec serializes them
- fail the sensor and RTR-session-close paths on a 200 whose envelope carries
  only errors, and surface partial sensor errors
- give Delete Indicators its own filter so a stale alert query cannot widen it
- drop the pre-selected network-isolating host action

* fix(crowdstrike): correct the RTR command tier, IOC update contract, and US-3 region

Independent re-validation against gofalcon's swagger-generated models and
CrowdStrike's developer center turned up several wire-level errors.

- Real Time Response advertised "eventlog backup"/"export"/"list", "reg query",
  ifconfig, and users as base commands. base_command names a command family and
  subcommands belong in command_string; the eventlog write variants are Active
  Responder commands that would fail on scope under this Read-scoped tool, and
  ifconfig/users appear in neither authoritative list. The block now offers the
  16 documented read-tier families and the contract enforces them.
- Indicator updates accepted an entry with no id, which cannot name a record,
  and accepted type/value, which the update model does not expose. Creates
  accepted an entry with no type, value, or applied_globally -- the one property
  CrowdStrike marks required, and the one that decides fleet-wide scope.
- CrowdStrike documents that PATCH overwrites any omitted field with a blank
  value. The contract can only catch blanks, so the update tool now tells the
  caller to read the indicator first and resend its full field set.
- Added the US-3 commercial region, which was missing from every cloud list.
- Aggregate queries silently dropped percents and filters_spec.
- Deleted the response-envelope body unwrap: no endpoint this integration calls
  returns that shape, and getFalconErrorMessage never honored it anyway.
- Softened the Detects and Incidents claims to what the sources actually state.

A tool description longer than the docs generator's 600-character id-search
window silently publishes as an empty string; three descriptions had crossed it.
Shortened them and added a test that fails before the catalog goes blank.

* docs(crowdstrike): name the endpoint and Identity Protection scope on the sensor tools

The three sensor tools were the only ones in the family that named neither their
endpoint nor their OAuth2 scope, and none of them said these are the domain
controllers Falcon Identity Protection monitors rather than Falcon endpoint
sensors -- a distinction an agent choosing between them and the Hosts tools has
no other way to make. Identity Protection Entities: Read is also a separate
product entitlement from Hosts and Alerts.

* refactor(crowdstrike): say which ID caps are CrowdStrike's and which are Sim's

Every bulk-ID limit claimed CrowdStrike as its source, but only the sensor
(5000), host action (100), indicator batch (200), and Spotlight (400) caps are
published. The alert, host group, indicator, and case caps are Sim's own bound
on request size, and the validation message now says so instead of attributing
a limit CrowdStrike does not document.
@waleedlatif1
fix(workflow): derive the webhook URL only where a sub-block shows one (
57611bd 
#6758)

`sub-block.tsx` mounts `useWebhookManagement` for every sub-block in the editor
panel, and `getBaseUrl()` throws when NEXT_PUBLIC_APP_URL reads empty, so a
missing deployment value took down the whole workflow route instead of the one
webhook field. The hook already gates its query and store writes on
`useWebhookUrl`; the URL now agrees.
@waleedlatif1
feat(servicenow): semantic incident, change, catalog, approval, CMDB,…
9e67655 
… and knowledge tools (#6747)

* feat(servicenow): add semantic incident, change, catalog, approval, CMDB, knowledge, and directory tools

The ServiceNow block only exposed generic Table API CRUD, so every real task
started with "which table is that on?". This adds 27 semantic tools that wrap
the same Table API plumbing under the names customers actually use.

- Incidents: create, get by number or sys_id, search, update, resolve, close,
  and append a work note or customer-visible comment.
- Change: create, get, list, update, move state, and list change tasks through
  the documented Change Management API.
- Service catalog: browse items, order one via the Service Catalog API
  order_now endpoint, and list or get requested items.
- Approvals: list pending approvals for an approver, approve, and reject.
- CMDB: search CIs on any class, read a CI with its inbound and outbound
  relations through the CMDB Instance API, and list cmdb_rel_ci rows.
- Knowledge: search and read articles through the Knowledge Management API.
- Directory: find a user by email or user name and list group members, which
  is what fills assigned_to and assignment_group.

Reference fields are the usual source of confusion, so every semantic read
defaults to sysparm_display_value=all — a reference comes back as both its
sys_id and its label — and every semantic write exposes
sysparm_input_display_value so a display name can be written instead of a
sys_id. Coded state values are exposed as labelled dropdowns built from one
constants module rather than raw integers.

The shared instance-URL, Basic Auth, sysparm, envelope, and error handling now
live in tools/servicenow/utils.ts, and the existing eight generic tools were
moved onto it rather than keeping their own copies.

* fix(servicenow): stop per-operation subblock defaults colliding on a shared id

Subblock initial values are seeded into block state keyed by subblock id, so
two subblocks sharing an id leave one stored value and the last definition
wins. Three ids were duplicated with differing defaults:

- `displayValue` was defined twice, unset for the generic Table API tools and
  `all` for the semantic ones. The semantic definition won, so a new block set
  to Read Records or Aggregate Records sent `sysparm_display_value=all` — a
  wire change to two already-shipped tools.
- `state` was defined four times. The Approval State definition won, so every
  new block carried `state=requested`, which Create Incident wrote to the
  incident and Move Change State used instead of its own `-5` default.

Give the colliding controls their own ids and map them back to the tool params
per operation, so the generic tools keep their original request shape and each
semantic operation keeps its own default.

Also correct descriptions that overstated what the API does: the LIKE operator
is not documented as case-sensitive, List Requested Items has no requester
filter, and the Change Management API task shape differs from the Table API.

Adds tool tests covering the refactor invariants for the eight pre-existing
Table API tools and the display-value separation.

* feat(servicenow): read a change request's real next states from the instance

The change tools describe state transitions using the base-system codes, which
only hold on an instance that has not customized its change model. ServiceNow
publishes an endpoint that answers the question directly for the record in
hand, so use it rather than keep assuming.

GET /api/sn_chg_rest/change/{sys_id}/nextstates returns the states reachable
from the change request, the instance's own state-value-to-label map, and, for
model-driven changes, each transition with the conditions it has and has not
met. The tool flattens the per-target-state grouping ServiceNow returns (each
transition already carries from_state and to_state, so nothing is lost) and
derives the states whose conditions currently pass.

Also record the sourcing for the coded values in constants.ts: the change
states and close codes are published as a table, but the incident state codes
are not — only 6 (Resolved) appears in the docs — so mark the rest as defaults
rather than guarantees. Note that sysparm_input_display_value also reinterprets
date and time values in the caller's timezone instead of GMT, which matters for
the change start and end dates.

* docs(servicenow): stop asserting undocumented coded values in placeholders

The additional-fields examples used hold_reason with a coded value of "1".
ServiceNow documents the On hold reason choices by label only — Awaiting
Caller, Awaiting Change, Awaiting Problem, Awaiting Vendor — and publishes
neither the column name nor the codes, so the example was asserting something
unsourced. Use a field whose value is caller-supplied instead, and record the
On Hold requirement on the incident state control using the labels the docs
actually give, including that Awaiting Caller makes Additional Comments
mandatory.

* fix(servicenow): drop phantom parent fields from the catalog order output

order_catalog_item read parent_id and parent_table off the order_now response.
Those fields belong to submit_producer, a different Service Catalog endpoint;
the documented order_now result is sys_id, number, request_number, request_id,
and table. Both outputs were therefore always null.

* fix(servicenow): correct what knowledge search returns as an article id

Search results carry a table-prefixed identifier — "kb_knowledge:9e528db1..."
— not a bare sys_id, while GET /knowledge/articles/{id} accepts only a bare
sys_id or a KB number. The output described it as a sys_id and the tool
description told callers it was what they needed to fetch the article, so
chaining the two tools on that field would fail. Point callers at the KB
number instead. Relevancy score is documented as a number, not a string.

* docs(servicenow): cite the page that actually documents approval statuses

The approval state constants pointed at the classic-approvals landing page,
which does not list the statuses. Approval status is documented separately and
names four — Requested, Approved, Rejected, and Not Requested.

* fix(servicenow): stop constant interpolation leaking into tool descriptions

The docs generator and the client-facing integration catalog read tool
descriptions from source rather than from the evaluated module, so a
template literal like `state ${INCIDENT_STATE.RESOLVED}` shipped to users
verbatim: `apps/sim/lib/integrations/integrations.json` and the published
ServiceNow integration page both rendered `${INCIDENT_STATE.RESOLVED}`
instead of `6`. Inline the base-system coded values in the description
text; the constants stay in use everywhere behavior depends on them.

Also drops an escaped `\'` in the `inputDisplayValue` description for the
same reason, and adds a standing guard test asserting no subBlock id
carries two different seeded defaults — the invariant behind the
per-operation defaulting bug, now checked structurally rather than only
through the four per-operation cases.

* refactor(servicenow): type the shared response boundary instead of any

`parseServiceNowResponse` returned `any`, so every tool reading `data.result`
did unchecked property access — a shape change on the instance side would have
produced a wrong-typed output silently rather than a type error.

Introduces `ServiceNowEnvelope` (`result?: unknown`) as the parser's return
type and narrows the record index signatures from `any` to `unknown`. Adds
`toRecordObject`, `readString`, and `readNestedNumber` so the tools that read
individual fields narrow deliberately at the point of use.

This surfaced five genuinely unchecked reads: Order Catalog Item, Get Knowledge
Article, and Search Knowledge were declaring `string | null` / `number | null`
outputs while emitting whatever the instance sent, and Get Change Next States
assigned an unvalidated object to `Record<string, string>`. Each now coerces or
drops a non-matching value rather than passing it through.

* fix(servicenow): publish the shared tool params and stop offering inert controls

The docs generator reads tool source rather than importing it, so the shared
`params.ts` consts the semantic tools spread were dropped from every published
Input table — 27 of 35 ServiceNow tools listed no instance URL, username, or
password at all. Follow a spread into the module it is imported from so those
rows are published; ten other integrations gain the rows they were missing for
the same reason.

Two controls were dead on arrival: Additional Fields was offered on Move Change
State and Add Incident Comment, and neither tool read it. Wire it through the
change transition, which needs it, and drop it from the comment tool, whose body
is exactly one journal field.

Every coded-value control was a select-only dropdown, so a customized instance's
state or close code was unreachable — sharpest on Move Change State, whose
target state is required and whose real codes come from Get Change Next States.
Make them comboboxes.

Also correct two doc claims ServiceNow does not publish (the incident state
citation pointed at a page that does not exist and compares the legacy
incident_state field; closing an incident is not documented as requiring
itil_admin), replace Record<string, any> with checked narrowing that surfaced
two unsound widenings, and document that List Change Tasks returns a fixed
{value, display_value} shape under `tasks` rather than `records`.

* fix(servicenow): stop one subblock id from carrying two value spaces

Subblock values are stored per block keyed by id, so an id reused across
operations keeps its value when the operation changes. Incident and change
shared `state`, and `closeCode`, `closeNotes`, `comments`, and the knowledge
search phrase were each reused for a different value space — so an incident
state could be written onto a change request, an incident close code sent as a
change close code, or an encoded query searched as knowledge text.

Give each value space its own subblock and republish it to the tool param from
the operation that owns it, the way targetState and approvalState already work.
The generic Table API ids stay exactly as they are, since renaming one would
orphan the stored value of every workflow already using those shipped tools.

The previous guard only compared seeded defaults, which is why this class stayed
hidden; the new one asserts against the merged params a tool actually receives.

* fix(servicenow): point the canvas sentences at the renamed subblocks

The split of the colliding subblock ids left the operation sentences anchored on
ids that no longer exist, so those clauses would silently drop from the card.

* fix(servicenow): validate collection members and split the fields projection

toRecordArray cast every member of a successful response, so a null or scalar in
a collection was handed to the next block as a record while the tool reported
success and its declared output said that could not happen. Members that are not
plain objects are now dropped, and knowledge articles and change transitions get
the same narrowing. The two response types that described an unverified inner
shape now say what is actually checked.

The 'fields' subblock also carried two value spaces: a JSON body on Create and
Update Record, a comma-separated projection everywhere else. Operations added
since read a separate returnFields control, so a body can no longer arrive as a
projection or the reverse. The shipped ids are untouched, since renaming one
orphans the stored value of every workflow already using those tools.
@icecrasher321
fix(forking): hide satisfied dependent configuration (#6723)
be20df9 
* fix(forking): hide satisfied dependent configuration

* fix(forking): keep dependent chains configurable

* fix(forking): invalidate stale dependent selectors
@waleedlatif1
feat(datadog): extend to 40 tools and align every operation with the …
cabd2e2 
…published OpenAPI specs (#6745)

* feat(datadog): add incidents, SLOs, dashboards, synthetics, Cloud SIEM, and APM tools

Extends the Datadog block from 12 to 39 operations, all verified against
Datadog's published OpenAPI specs:

- Incidents (v2, public beta): list, get, create, update, add todo
- SLOs (v1): list, get, create, update, delete, history
- Dashboards (v1): list, get, create, delete
- Synthetics (v1): list tests, get test, latest results, trigger, pause/resume
- Cloud SIEM (v2): search signals, get signal, update triage state, assign,
  list detection rules
- APM: search spans (v2), list Service Catalog definitions (v2)

Adds tools/datadog/utils.ts so every tool builds its URL from the configured
site/region and shares the JSON:API-aware error extraction, and handles the
v1 flat vs v2 envelope shapes and cursor pagination per endpoint.

* fix(datadog): align every operation with the published OpenAPI specs

Validated all 39 shipped operations (plus the 12 pre-existing ones that had
never been spec-checked) against the DataDog v1 and v2 OpenAPI schemas.

- `POST /api/v2/downtime` requires `monitor_identifier`, so a downtime created
  without a monitor id was rejected. Default to the `*` monitor tag.
- A one-time downtime schedule declares `additionalProperties: false` and
  accepts only `start`/`end`; the timezone moves to `display_timezone`.
- `GET /api/v2/downtime` has no `monitor_id` filter, and the response carries
  no `disabled` attribute. Downtime ids are UUID strings, not numbers.
- Drop scaffold types for operations that do not exist (metric metadata, event
  query, monitor update/delete/unmute, host listing) along with their fields.
- Note that monitor mute is no longer published in the v1 specification.
- Add browser Synthetic test results, which the browser-specific endpoint
  returns with its own camelCase step-count shape.
- Replace every `any` with a spec-derived interface, keeping the polymorphic
  service-definition schema opaque.

* fix(datadog): remove remaining any types and declare every returned output field

Replace the six surviving `Record<string, any>` request-body and response-cast
sites with concrete spec-derived shapes, and declare the output fields that
transformResponse already returned but outputs omitted:

- create_downtime / list_downtimes: timezone, created, modified
- create_monitor / get_monitor: options, creator
- list_monitors: message, priority, options, created, modified, creator
- query_logs: content.attributes, content.tags
- update_security_signal_state / _assignee: type; assignee also gained the
  archiveReason/archiveComment pair its sibling already declared
- query_timeseries: series gained the items shape it never described

* fix(datadog): stop dropping downtime targeting inputs in the block mapping

create_downtime accepts monitorTags, timezone and muteFirstRecoveryNotification,
but the block exposed no inputs for them and never forwarded them. Monitor-tag
targeting silently fell back to the `*` tag, so a downtime meant for one team's
monitors muted every monitor in scope. Adds the three advanced sub-blocks and
wires them through.

Also routes list_downtimes' currentOnly through toSwitchBoolean. A switch yields
the strings 'true'/'false', and 'false' is truthy, so turning the toggle off
still sent current_only=true. Every other switch in the block already used the
helper; this was the last raw one.

* fix(datadog): correct metric type codes, stop SLO update data loss, drop unpublished mute

Independent re-validation of all 39 operations against the DataDog/datadog-api-client-go
generator specs (v1 and v2 openapi.yaml) rather than the client-rendered docs site.

Correctness:
- submit_metrics sent inverted MetricIntakeType codes (gauge as 0/unspecified, rate as 1/count,
  count as 2/rate), silently changing how Datadog aggregated every submitted series. The spec
  enum is 0 unspecified, 1 count, 2 rate, 3 gauge; an unrecognized type is now omitted so
  Datadog infers it. Also stops stamping an invented `resources: [{name:'host'}]` default and
  now forwards `interval`, which Datadog requires for count and rate metrics.
- update_slo replaced the whole SLO with only the fields the caller filled in, so editing one
  field erased description, tags, query, monitor_ids, groups, thresholds, and timeframe.
  PUT /api/v1/slo/{slo_id} is a full replacement, so the stored SLO is now read first and the
  supplied edits are overlaid onto it, with the read-only fields stripped.
- update_incident admitted empty strings, so a blank input could blank a stored incident title
  or fail as an invalid date-time.
- query_timeseries reported a failed query as success: Datadog returns 200 with a non-ok
  `status` and the reason in `error`.
- create_monitor swallowed malformed options JSON and created a monitor with no thresholds.
- send_logs rebuilt each entry from a fixed field list, discarding the custom attributes
  Datadog accepts as additionalProperties, and padded absent optional fields with empty strings.

Removed:
- mute_monitor. /api/v1/monitor/{monitor_id}/mute is absent from the v1 spec entirely, there is
  no unmute counterpart to reverse it, and downtimes are the supported mechanism.

Contract accuracy:
- Security signal search advertised relative times ("now-1h"); the spec types filter.from/to as
  format: date-time. Descriptions, placeholders, and wand prompts now produce ISO-8601.
- list_incidents advertised an `include` value ("integrations") that is not in the spec enum,
  and neither incident tool trimmed the comma-separated list, so "users, attachments" 400d.
- Invalid "ok" group state dropped from both monitor descriptions.
- time_slice removed from SLO create input, which cannot build one without an SLI specification.
- DatadogSite gains ap2, uk1, and us2.ddog-gov.com.

Pagination and errors:
- list_downtimes silently truncated at Datadog's default 30 with no way to page; adds
  page[limit]/page[offset] and surfaces totalCount.
- query_logs returned a cursor it had no way to accept back.
- Error extraction consolidated onto datadogErrorMessage, which now also reads the
  dictionary-shaped errors of the SLO delete conflict. Ten tools were reading `.detail` off
  plain strings or the raw entry off objects, degrading every failure to a bare status line.
- Debug logging removed from list_monitors.

Adds 29 regression tests, each verified to fail when its fix is reverted.

* fix(datadog): add SEV-0, document page-size caps, drop unsourced output defaults

- The severity dropdown omitted SEV-0, which IncidentSeverity allows and both incident
  tool descriptions already advertised.
- Page-size descriptions now state Datadog's documented default of 10 and cap of 100
  instead of an arbitrary example, so an agent does not request an out-of-range page.
- trigger_synthetics_tests emitted an explicit null for a string-typed optional output,
  and update_synthetics_status reported 'live' on the error path regardless of what the
  caller actually requested.

* fix(datadog): keep mute_monitor and add the missing unmute counterpart

Reverses the removal in the previous commit. Absence from the datadog-api-client-go
generator spec showed the endpoint is unpublished there, not that it is retired:
Datadog's official Python client still implements it on master as
`Monitor.mute(id, scope=, end=)` and `Monitor.unmute(id, scope=, all_scopes=)`
(datadogpy datadog/api/monitors.py), which `_trigger_class_action` resolves to
`POST /api/v1/monitor/{id}/mute` and `/unmute` with exactly those body fields.

mute_monitor has also been in the block since #2175 in December, so dropping it would
have broken existing workflows for an endpoint that two independent sources agree is live.

The genuine defect was that muting was a one-way trapdoor: Sim could mute a monitor but
had no way to reverse it. Adds datadog_unmute_monitor, sharing the monitor ID and scope
inputs with mute, so the operation is recoverable from the same block.

Also: mute no longer discards the response body (it now reports the monitor id, name, and
state), routes errors through datadogErrorMessage, encodes the monitor ID in the path, and
stops dropping an explicit `end` of 0.

* fix(datadog): make downtime targeting explicit and reach downtime pagination from the block

Addresses the review findings on the previous round.

- create_downtime accepted both a monitor ID and monitor tags but `monitor_identifier` is a
  oneOf, so it silently kept the ID and dropped the tags, muting a different set of monitors
  than the caller asked for. It now rejects the ambiguous combination.
- create_downtime ran Number.parseInt on the monitor ID with no validation, so a non-numeric
  value became NaN and serialized as null inside monitor_identifier. It now uses the same
  parseMonitorIds guard the SLO path already had, naming the offending value.
- list_downtimes gained limit/offset in the tool but the block exposed neither, so no
  block-driven call could page past Datadog's default. Adds the two sub-blocks and wires them
  through the params mapper.
- The block did not declare the totalCount the tool now returns, so nothing downstream could
  bind to it.

* fix(datadog): tolerate non-string list inputs and keep the shipped mute subblock ids

Both defects were introduced by this branch.

- splitCommaList called .split on its argument, so routing create_downtime's monitorId
  through it turned a legitimate numeric input into a TypeError before the request was
  built. A <Block.output> reference to get_monitor or list_monitors resolves to a number,
  and an LLM tool call can pass a number or an array, so the helper now normalizes all
  three shapes. The previous Number.parseInt path had accepted a number by coercion.
- Adding the unmute operation renamed the mute subblock ids scope/end to muteScope/muteEnd.
  Workflow state is persisted by subblock id, so every existing Mute Monitor block would
  have kept the old keys and silently lost its scope and end time. Restored the shipped ids;
  both are still unique block-wide and no operation reads another operation's value.

* fix(datadog): compare downtime targets after parsing, not before

A whitespace-only Monitor ID is truthy as a raw string but parses to no monitor, so
the oneOf conflict guard rejected a valid tag-targeted downtime whenever the untouched
Monitor ID field carried blank text. Both sides are now compared after parsing.
@waleedlatif1
feat(cloudflare): add WAF rulesets, rate limiting, Zero Trust Access,…
8a44621 
… R2, Workers, and Tunnels (#6740)

* feat(cloudflare): add WAF rulesets, rate limiting, Zero Trust Access, R2, Workers, and Tunnels

Extends the Cloudflare integration past DNS/zones/cache with the security and
Zero Trust surface:

- Rulesets engine (zone-scoped): list rulesets, get a ruleset, read a phase
  entry point, and create/update/delete rules. WAF managed-rule overrides are
  surfaced through the http_request_firewall_managed entry point, since
  Cloudflare has no dedicated overrides endpoint.
- Rate limiting (zone-scoped) via the current Rulesets-based http_ratelimit
  phase, not the deprecated rate_limits endpoint.
- Cloudflare Access (account-scoped): applications, application policies,
  groups, identity providers, and service tokens.
- R2 buckets, Workers scripts/routes, and cloudflared Tunnels.

Destructive operations (delete application, delete policy, revoke service
token, delete rule, delete bucket) spell out their blast radius, and every
tool branches on the envelope's success flag rather than the HTTP status.

Security events are intentionally omitted: Cloudflare exposes them only
through the GraphQL firewallEventsAdaptive dataset, whose field list is not
documented outside schema introspection.

* fix(cloudflare): correct docs drift and remove any from the tool layer

Validation pass over all 47 Cloudflare tools against developers.cloudflare.com.

- Two tool descriptions still escaped a quote as \'. That reaches the model
  verbatim and truncates the generated MDX cell — the get_zone_settings
  `value` output row was missing from the published docs entirely. Both are
  now template literals, and the row is back.
- list_rulesets ignored pagination. The endpoint pages by cursor via
  result_info.cursors.after (not page/per_page), so a zone with many rulesets
  silently truncated with no way to page. Expose per_page + cursor and return
  the next cursor.
- The managed-ruleset override description claimed action and enabled were
  the overridable properties. They are the ones the Rulesets engine documents
  at every level, but individual managed rulesets add more: an OWASP Core
  Ruleset rule override also takes score_threshold. Corrected in both the
  tool output description and the block's action-parameters wand prompt.
  (sensitivity_level is a DDoS override, not a WAF one — deliberately absent.)
- list_tunnels/get_tunnel dropped the documented `metadata` field.
- list_r2_buckets appended order=name whenever any filter was set. `order`
  only qualifies `direction`, and `name` is its sole documented value.
- Path-interpolated IDs are trimmed, so a pasted ID with trailing whitespace
  no longer 404s.
- Replaced every `any` in the integration with checked types: a shared
  CloudflareEnvelope plus per-resource raw payload interfaces, read through
  readCloudflareResponse. The mappers in utils.ts were the widest hole —
  typing them caught four real output-shape mismatches (identity provider
  read_only, service token enabled, DNS record meta/priority, certificate
  geo_restrictions) that `any` had been hiding.
- BlockMeta only described DNS and zone work. Added templates and skills for
  the WAF, rate limiting, and Zero Trust Access surfaces the block now has.

Confirmed against the docs and left unchanged: rulesets/rate limiting are
zone-scoped and Access/R2/Workers scripts/Tunnels are account-scoped while
Workers routes are zone-scoped; tunnels live under /accounts/{id}/cfd_tunnel;
the ratelimit object is a sibling of action/expression, not nested in
action_parameters; every rate limiting period and mitigation_timeout option
matches the documented set; R2 delete returns an empty result so echoing the
requested bucket name is correct; app-nested Access policy endpoints are
current, not deprecated; and every tool fails on a 200 carrying success:false.

* fix(cloudflare): stop per-operation subblock defaults colliding on a shared id

Subblock initial values are seeded into block state keyed by subblock id —
both stores/workflows/utils.ts and lib/workflows/defaults.ts assign
`subBlocks[subBlock.id]` in a plain forEach — so two controls sharing an id
leave one stored value and the last definition in file order wins. Four ids
were duplicated with differing defaults:

- `type` was defined four times. The Access "Application Type" control is
  last, so every new block seeded `type = 'self_hosted'` and the three DNS
  record controls inherited it — Create DNS Record sent a Zero Trust
  application type as its record type. The subblock added on this branch
  broke a default on tools that shipped long before it.
- `status` was defined three times. The empty tunnel filter is last, so
  List Certificates lost its `all` default.
- `proxied` was defined three times. An empty filter is last, so Create DNS
  Record lost its explicit `false`.
- `action` was defined twice. The rate limiting dropdown is last, so the
  ruleset-rule action input was seeded `block`, quietly making "block live
  traffic" the default for a WAF custom rule the user never configured.

Give the colliding controls their own ids and map them back to the tool
params per operation, ahead of the coercions that read them, so each
operation keeps its own default. The other 17 duplicated ids agree on their
value and are left shared.

Adds tests covering each separated default plus a sweep asserting no id
carries two different seeded values, so a future duplicate goes red.

* fix(cloudflare): generate array include rules and allow bootstrapping a phase ruleset

The Access policy include wand asked for a JSON object while the tool parses
the field with parseJsonArrayParam, so generated rules failed validation.
Switch it to json-array, whose prompt reinforcement omits the object braces.

Rate limiting and WAF custom rules could only be appended to an existing
ruleset, but a zone that has never had a rule in a phase has no entry point
ruleset and returns 404, leaving no way to add the first rule. Add
cloudflare_create_ruleset for the documented POST /zones/{id}/rulesets
bootstrap, seeded with optional initial rules.

* fix(cloudflare): correct verified API defects and stop filters leaking into writes

Independent re-validation of all 48 tools against developers.cloudflare.com
turned up defects that the shipped tools would have hit on their happy path.

Delete DNS record reported every success as a failure. That endpoint is the
one Cloudflare v4 response with no envelope — its documented body is
`{"result":{"id":...}}` with no `success` — so `!data.success` was always
true. Branch on an explicit `=== false` instead.

The two replace-semantics PATCH endpoints could silently destroy live config.
Update rate limit rule defaulted a missing action to `block`, converting an
existing `log` or challenge rule into a hard block on real traffic; update
ruleset rule left action and expression optional and had no `ratelimit` or
`logging` passthrough, so updating a rate limiting rule stopped it rate
limiting. Both now require the fields the replacement needs, and the ruleset
rule carries the two nested objects through.

Access applications were unbuildable for most types: `domain` was required,
but it does not exist on the saas, app_launcher, warp, biso, dash_sso,
infrastructure, mcp, mcp_portal, or proxy_endpoint request variants. The
application type enum was also six values behind. Access group `is_default`
is an array of rule objects, not a boolean.

Purge cache merged every supplied target into one body, but the purge body is
a one-of over the five target kinds; it now names the conflict instead.

The remaining fixes are documentation drift: the priority field is MX and URI
only (an SRV record carries priority inside its content), the certificate
status filter documents only "all", the Worker tag filter takes tag:allowed
pairs, and the managed-rule override list conflated the DDoS-only
sensitivity_level with the WAF rule-level set.

Separately, controls that share a subBlock id share one stored value, and
`shouldSerializeSubBlock` short-circuits on `mode: 'advanced'` before it
evaluates `condition` — so a hidden list filter was reaching a write. A
`list_dns_records` content filter could overwrite a record's content, cache
tags could be written onto a DNS record, and the zone status enum could reach
the tunnel list, whose enum is disjoint. Filters that differ from the value
they collided with now carry their own id, remapped through one table before
any coercion. Sharings that mean the same thing everywhere are unchanged.

Aliases are cleared by explicit assignment rather than destructuring, because
the executor merges the mapper's output over the raw inputs and a merely
omitted key survives as its raw subBlock string. The tests assert on that
merged result, and three mechanical invariants now go red on a new collision:
no id spans a read filter and a written value, no dropdown id carries two
option sets, and no hidden advanced control feeds an operation that cannot
render it. That last one found the name filter reaching three list operations.

* docs(cloudflare): point self_hosted_domains at its replacement

Cloudflare deprecated the field in favour of destinations, which the tools
already surface. The output stays — Cloudflare still returns it — but the
description now says which one to read.

* refactor(cloudflare): drop a dead exception from the empty-type guard

create_dns_record now takes its record type from the recordType control,
whose dropdown has no empty option, so the operation can never reach this
guard with an empty type. Clear it unconditionally.

* fix(cloudflare): point the canvas sentences at the renamed filter controls

The list filters that were split off their write-side twin kept their old ids
in canvasPresentation, so seven clauses referenced a control that is no longer
visible for that operation — check:canvas-sentences catches exactly this, and
a broken clause fails silently on the card rather than throwing.

* fix(cloudflare): stop the rate limiting action defaulting on a replacing update

Making action required on update_rate_limit_rule was only half the fix: the
Action dropdown still seeded block for the update operation too, so an update
that edited only the threshold kept sending block and converted a live log or
challenge rule into a hard block — exactly the harm the required flag was
meant to prevent. The update now has its own control with no seeded value, so
the action is something the caller states rather than inherits.

The certificate status filter also still offered Active and Pending, which
Cloudflare does not document for that endpoint; the only documented value is
all, and omitting it returns active packs.

* fix(cloudflare): stop the Access replacements seeding a type and a decision

Same class as the rate limiting action: both Access updates are full
replacements, and the shared controls seeded self_hosted and allow for the
update operations too. Editing only a policy's include rules would silently
convert a live deny, bypass, or non_identity policy to allow — widening who
gets in — and editing an application would rewrite what it IS.

Each update now has its own required control with no seeded value, so the
type and the decision are stated rather than inherited. Regression tests
cover both, and the canvas sentence follows the renamed decision control.
@waleedlatif1
fix(sandbox): undefine the raw fetch host bridge before user code runs (
6e5c337 
#6761)

* fix(sandbox): undefine the raw fetch host bridge before user code runs

* test(sandbox): scope the hardening assertions to each execution path

* fix(sandbox): preserve the fetch global's property attributes
@j15z
feat(tooling): enforce docs freshness and modernize agent skills (#6756)
7f936dc 
* feat(docs): fail CI when generated integration docs are stale

* fix(docs): don't flag delete-then-recreated trigger pages in check mode

* docs(skills): require docs:check in the integration authoring skills

* chore(skills): migrate agent commands to native skills

* fix(skills): clean orphaned Claude projections
@waleedlatif1
feat(mssql): add Microsoft SQL Server integration (#6739)
6a29a9e 
* feat(mssql): add Microsoft SQL Server integration

Add a Microsoft SQL Server block backed by six tools (query, execute,
insert, update, delete, introspect), mirroring the existing PostgreSQL
and MySQL integrations.

Connections go through the `mssql` (Tedious) driver: `connectionTimeout`
is top-level while `encrypt`, `trustServerCertificate`, and
`instanceName` live under `options`, and `port` is omitted when a named
instance is used. Values are bound as `@paramN` via `request.input()`;
no user value is interpolated into SQL. Identifiers are bracket-quoted
after validation and WHERE clauses run through the shared injection
guard.

Introspection reads INFORMATION_SCHEMA plus the `sys.indexes` catalog
views for tables, columns, primary keys, foreign keys, and indexes.

The icon is a placeholder database cylinder drawn with `currentColor`
until the real brand mark lands.

Requires `bun install` for the new `mssql` / `@types/mssql` deps.

* feat(mssql): use the SQL Server brand mark on a white tile

* fix(mssql): pin the validated IP and correct the introspection catalog reads

Tedious exposes `options.connector`, a hook that replaces its own
resolve-and-connect path, so the connection can be pinned to the address
`validateDatabaseHost` already approved instead of re-resolving the
hostname. `server` stays the hostname because tedious derives the TLS
`servername` from it independently of the connector, so SNI and
certificate validation survive the pin. This brings MSSQL in line with
the PostgreSQL and MySQL tools.

Named instances are dropped: tedious resolves them with a UDP SQL Server
Browser lookup issued outside the connector, and node-mssql deletes
`port` whenever `instanceName` is set, so no configuration leaves a
named instance pinned. A named instance is reachable through its static
TCP port.

Introspection fixes:
- index key columns now filter on `key_ordinal > 0`; INCLUDEd columns
  and partitioning columns both report `0` and were being returned as
  key columns, ordered ahead of the real ones
- foreign keys resolve through `sys.foreign_keys` /
  `sys.foreign_key_columns` rather than
  `INFORMATION_SCHEMA.REFERENTIAL_CONSTRAINTS`, whose join to
  `TABLE_CONSTRAINTS` has no row when a key references a unique index
  and so dropped the key entirely
- `is_unique` is a `bit`, which tedious maps to a boolean, so it is
  coerced rather than compared
- schemas come from `sys.schemas`, which needs only `public` and carries
  no metadata-visibility caveat

The WHERE-clause guard also covers `WAITFOR TIME`, `OPENQUERY`,
`OPENXML`, the legacy `master..sys*` compatibility views, and extended
and OLE-automation procedures beyond `xp_cmdshell`.

Regenerates the docs and catalog artifacts the icon change left stale.

* chore(mssql): commit the lockfile entries for mssql and its tedious dependency tree

* fix(mssql): make the Query operation genuinely read-only

The block label, tool description, and docs all present Query as SELECT-only
while the route ran whatever T-SQL it was given, so an agent picking
mssql_query because "it is only a SELECT" could delete rows. Screen the
statement for mutating keywords with string literals stripped, which also
catches the WITH ... DELETE form that a leading-token check would miss.

Also switch the tool barrel to absolute imports per the repo convention.

* fix(mssql): compose the shared WHERE guard and close the semicolon-less batch gap

The local validateWhereClause re-derived an older copy of the shared patterns
and scanned raw text, so it missed a bare 1=1 and false-positived on prose in a
quoted value. Delegate to validateSqlWhereClause, which masks string literals
first, and keep only the SQL Server surfaces it has no reason to know about.

T-SQL needs no statement terminator, so every semicolon-anchored stacked-query
check reads straight past `id = 1 DROP TABLE dbo.users`. Screen for a bare
statement-introducing keyword to close that; word boundaries leave ordinary
column names like updated_at and deleted_at untouched.

Export maskSqlStringLiterals so the dialect layer masks the same way the shared
guard does rather than carrying a weaker single-quote-only copy.

* fix(mssql): screen administrative T-SQL and reject batches in the read-only path

The previous round left two keyword lists maintained separately, and both were
short: DBCC, KILL, CHECKPOINT, USE, and DENY were in neither, so
`SELECT 1; DBCC SHRINKDATABASE(...)` and `id = 1 DBCC SHRINKDATABASE(...)`
both got through. Collapse them into one MSSQL_STATEMENT_KEYWORDS shared by the
query and WHERE screens so a keyword cannot be covered in one place and missed
in the other, and add the administrative commands.

Also reject any second statement after a semicolon in the Query path outright.
That closes SELECT 1; <anything> structurally instead of by naming the anything,
so the blacklist no longer has to be exhaustive to hold.

* fix(mssql): reject SQL comments in the read-only query path

A block comment placed inside a keyword splits it as far as a lexical scan is
concerned, so keyword coverage cannot settle whether the server rejoins the
halves. Refuse comments in the Query path instead of modelling the tokenizer.
A SELECT sent through this operation has no need for one, and Execute Raw SQL
still accepts them. Masking leaves comment markers intact, so a literal
containing -- still passes.

* fix(mssql): close the masker-desync bypasses and correct the catalog reads

Every T-SQL screen runs over the shared literal masker, which was written for
the MySQL dialect. Three ways to desynchronise it let real SQL hide inside what
the masker believes is a string, two of which survived the existing even-quote
check:

- a backslash before a quote. T-SQL has no backslash escape, so the server
  closes the literal where the masker swallowed the quote and runs the rest as
  code. `a='x\' DELETE FROM dbo.t WHERE b='y'` holds four quotes and masks the
  DELETE out of the keyword screen entirely, so the read-only Query operation
  would run it.
- a double quote inside a bracketed identifier, which the bracket rule missed
  because it only looked for single quotes.
- any unbalanced double quote or backtick, which the parity check did not cover.

All three now fail closed. Introspection also filters hypothetical and disabled
indexes, which were reported as if they were live, and resolves the referenced
side of a foreign key through sys.schemas so a cross-schema reference is no
longer an ambiguous bare table name. Values bound through request.input are
serialized when they are nested JSON, which the driver otherwise rejects with a
bare "Invalid string.".

* test(mssql): cover the block param merge and the operation-to-tool map

Asserts on the merged `{ ...inputs, ...buildParams(inputs) }` the generic tool
handler forwards rather than the mapper's return, since a key the mapper omits
keeps its raw subBlock value through that merge. Pins the TLS toggles to their
string form end to end — a switch subBlock would serialize `'false'`, which is
truthy, and the route contract would coerce the user's off into on — and checks
that duplicate subBlock ids agree on their seeded default.

* fix(mssql): release a pool whose connect failed, and allow a keyword with no trailing space

Only a pool handed back to the route reaches its `finally`, so a pool whose
connect rejected leaked its tarn resources — one per attempt when a bad
credential is retried. It now closes itself, and a failure to close cannot mask
the connect error the caller needs.

The read-only screen also anchored on `\s` after the opening keyword, which
refused valid reads like `SELECT*FROM dbo.users` and `SELECT(1)`. A word
boundary accepts those while still refusing `SELECTX`, and cannot loosen the
screen — the keyword and batch checks run over the whole statement regardless.

* chore(mssql): regenerate tool metadata and the integration catalog after rebase

Artifacts rebuilt with the generators rather than hand-merged, so they carry
both the mssql entries and the tools that landed on staging in parallel.

* fix(mssql): reject a parenthesised or negated constant tautology in a WHERE clause

The shared guard recognises `OR 1` but not `OR (1)`, `OR ((1))`, `OR NOT 0`, or
`OR NOT (FALSE)` — a parenthesis or a NOT between the operator and the constant
hides it. Both patterns require the constant to be the whole parenthesised term,
so a real disjunct such as `OR (1 = priority)` is untouched.

This narrows the gap rather than closing it, and is not meant to close it: an
always-true expression is not lexically decidable in general, which is why the
WHERE screen stays documented as defense-in-depth rather than a boundary.

* chore(mssql): regenerate tool metadata after rebase onto staging

Rebuilt with the generators so the artifacts carry the servicenow and
crowdstrike tools that landed on staging alongside the mssql entries.

* fix(mssql): screen trigger and state statements, and drop the space anchor on Execute

DISABLE and ENABLE were missing from the shared statement list, so
`SELECT 1 DISABLE TRIGGER dbo.audit ON dbo.users` passed the read-only screen as
a semicolon-less batch and turned auditing off. SET, BEGIN, COMMIT, and ROLLBACK
are added with them, since session and transaction state are reachable the same
way. FETCH is deliberately left out: OFFSET ... FETCH NEXT is the standard paging
clause, and screening it would reject the ordinary paged SELECT.

Execute Raw SQL anchored its allowlist on `\s`, which refused `EXEC(@SQL)` —
the ordinary form of dynamic SQL, on the one operation meant to run it. It now
uses `\b`, matching the read-only screen.
@waleedlatif1
docs(cli): add a CLI docs section generated from the command tree (#6762
fed891f 
)

* docs(cli): add a CLI section, generated from the command tree

The `sim` CLI shipped with no coverage in the docs site. Adds a fourth
top-level tab for it, and moves Academy last.

The command reference is generated. `sim` exposes 147 leaf commands across
33 groups, most of them derived at runtime from the v2 route contracts, so a
hand-written reference would be wrong the week after it was written. The
generator walks the command tree `buildProgram()` hands to commander — the
same tree the terminal parses — rather than re-deriving it from the contract,
which would be a second implementation free to describe commands nobody can
invoke. `check:cli-docs` is a zero-arg `check:*` script, so the existing audit
runner picks it up and stale pages fail CI.

Generating against the real tree surfaced a collision it had been hiding:
`bulkUpdateKnowledgeDocuments` and `updateKnowledgeDocument` both derived to
`sim knowledge documents update`. Commander resolves a duplicate to the first
match, so the bulk form shadowed the single-document one and its flags were
unreachable while still appearing in `--help`. The bulk form is now
`batch-update`, matching how `tables rows batch-delete`/`batch-update` already
handle the same REST overload, and the generator fails on any duplicate path
so the next one cannot land silently.

Five hand-written guides cover install, auth, configuration, output formats,
and scripting. Also corrects two commands in the package README that do not
exist as documented (`tables columns <tableId>`, and `--sort score:desc`,
which is JSON).

* docs(cli): document every flag from the contracts, add troubleshooting and a single-page reference

The command reference was structurally complete but said almost nothing: 223 of
377 flags rendered as "Set sort by" because the CLI only ever read flag help
from its own contract overrides, and fell back to restating the flag name.

The prose already existed. The v2 route contracts carry 931 `.describe()` calls
and the OpenAPI specs publish all of them — 327 parameters and 282 body
properties, 100% coverage — but the generated operation table dropped every one,
carrying only a per-operation summary. It now carries the field descriptions,
the path-parameter descriptions, and positional help, so `--help` and the docs
explain a flag the same way the API reference does. Placeholder descriptions are
now zero, and 147/147 commands, 377/377 flags and 130/130 arguments are
documented.

`check:cli-docs` fails on a request field with no `.describe()` rather than
letting it render as documentation that says nothing.

Also in this pass:

- Commands are root-level sidebar entries under a Commands heading rather than
  a folder, and headings are the command's description, so the table of
  contents distinguishes entries at the first word instead of repeating
  "sim knowledge documents …" fourteen times. A guard fails the build if two
  descriptions on a page collide, since they would share an anchor.
- A single-page `Complete reference` carrying all 147 commands, for in-page
  search and for agents fetching `/cli/reference.mdx`. It keys on exact command
  paths because descriptions are only unique within a group.
- A troubleshooting page, with every message copied from the source.
- Table columns are sized by a local component; the flag column was starved
  while descriptions kept most of the row empty.
- The prerelease install channels are dropped from the docs and the package
  README, which is what npm renders.

* fix(docs): match the CLI tab by path segment, and escape backslashes before pipes

`pathname.includes('/cli')` also matches `/integrations/clickup` and
`/integrations/clickhouse`, so both existing integration pages lit the CLI tab
and unlit Documentation. Matching is now per path segment. Anchoring to the
start would not work either — a non-default locale prefixes the path, as in
`/ja/cli` — so the segment is matched wherever it sits.

Table cells now double a backslash before escaping pipes. A value ending in one
turned `a\` + `|` into `a\\|`, which the table parser reads as an escaped
backslash followed by an unescaped pipe, splitting the cell early. Nothing in
the command surface contains a backslash today, so this was latent rather than
visible.

The reference page's global options table is two-column and was being wrapped in
`CommandTable`, which sizes the second column for the `Required` cell of the
three-column tables and crushed the description into 5.5rem. It now matches the
overview page, which leaves that table unsized.
@waleedlatif1
feat(microsoft_ad): licensing, security, audit, role, and device oper…
257029a 
…ations (#6742)

* feat(microsoft_ad): licensing, security, audit, role, and device operations

Deepens the Microsoft Entra ID block from 12 to 36 tools against the Microsoft
Graph v1.0 reference: license assignment and tenant SKUs, password set/reset,
sign-in session revocation, authentication methods, sign-in and directory audit
logs, app role and directory role assignments, service principals, device reads,
and conditional access policy reads.

Device write (device-update, device-delete) is deliberately excluded. Both
document Directory.AccessAsUser.All as their only delegated scope, with the
higher-privileged read documented as unavailable, so supporting them would mean
requesting tenant-wide act-as-the-user directory access for two operations that
additionally require the caller to hold Intune Administrator.

Also drops an undocumented ?$select= from create_user that was silently nulling
department and accountEnabled in the response.

* fix(microsoft_ad): resolve OData filter and search by owning operation

The params mapper assigned result.filter from each filter subBlock in turn, so
the last non-empty one won regardless of the selected operation. Because a
subBlock keeps its value after the operation changes, a filter written for one
endpoint was sent to every other collection operation — invalid OData against a
different Graph resource, or a silently wrong page.

Resolves the filter and search terms from an explicit operation-to-field map
instead, so each operation reads only the field it owns.

* fix(microsoft_ad): clear non-owning filter and search on the merged inputs

The executor merges { ...inputs, ...transformedParams }, so declining to copy a
stale filter is not enough — the serialized value survives the merge and still
reaches the tool. Advanced-mode subBlocks are serialized on non-emptiness alone
and never have their condition evaluated, so the value is present even when the
field is hidden.

Write filter and search on every operation, as undefined when the operation owns
neither, so the merge clears them.

* fix(microsoft_ad): clear the MFA flag and let paged user operations continue without a User ID

The set_password MFA dropdown only wrote its key when non-empty, so the "No Change"
empty string survived `{ ...inputs, ...transformedParams }` and reached Graph in place
of a boolean. Assign it explicitly, including as `undefined`, the same way `filter` and
`search` are handled.

`list_user_app_role_assignments` and `list_user_devices` page by `@odata.nextLink`, and
both tools already treat `userId` as optional once a continuation URL is supplied. Drop
them from the required set when Next Page is filled in so pagination-only runs pass block
validation.

Also note on the reset_password output that a generated password reaches workflow outputs,
run history, and the model, matching how other tools that return secrets document exposure.

* fix(microsoft_ad): require the service principal ID only on the first page

Every other single-resource ID field pairs its condition with a matching required
rule; servicePrincipalId had none, so a first-page run could pass block validation
with an empty ID and fail inside the tool instead. Require it unless a continuation
URL is supplied, matching the paged per-user operations.

* fix(microsoft_ad): reject a continuation URL from a different collection

Every paged operation reads the one shared Next Page field, and a subBlock keeps
its value after the operation changes. Paging /users and then switching the block
to /devices short-circuited back to the user page, silently returning the previous
collection instead of the selected one.

Assert the continuation URL's terminal path segment against the collection the tool
actually reads, which also rejects a nextLink pasted from an unrelated response.
@waleedlatif1
fix(docs): serve JSON-LD in the HTML, fix sidebar spacing, and tighte…
025ea4d 
…n the CLI guides (#6763)

* docs(cli): use -g for the install, and cut the prose that was not pulling weight

`--global` is valid but `-g` is what every comparable CLI documents, and the
long form only came from the package README. Also drops the yarn tab: it read
`yarn global add sim`, which works on Yarn 1 only — Yarn 2 removed global
installs, so that command fails for anyone on a modern Yarn. Adds `npx sim` for
running without installing.

The guides had accumulated design rationale that belongs in code comments rather
than user docs — why the filter grammar is JSON, why the config section naming
is asymmetric, why an unexpected error keeps its stack trace. Surveying how gh,
Vercel, Turborepo, Deno, Bun and Supabase write theirs, none carry that kind of
justification, and callouts are reserved for content whose absence produces a
wrong result rather than for general asides.

So: 1016 lines to 763, and 12 callouts to 3. The three that remain are the
pairing-code check, that `sim logout` does not revoke the key, and the
`--limit 100` default on `batch-delete`/`batch-update`, which silently truncates
a larger match. Troubleshooting drops the entries whose error message already
contained its own fix and keeps the seven whose cause is not obvious.

* fix(docs): render JSON-LD as native script tags so it reaches the HTML

All four structured-data blocks — WebSite, TechArticle, BreadcrumbList,
SoftwareApplication — were rendered with `next/script`, which never emitted a
script tag. Measured on a production build, `/api-reference/getting-started`
contained zero `<script type="application/ld+json">` elements; the payload
existed only in the `__next_s` client-injection queue and the RSC flight data,
so anything reading the served HTML saw no structured data at all. React was
also logging "Encountered a script tag while rendering React component" on every
page.

`next/script` is for loading and executing JavaScript. JSON-LD is data, and
Next's own guidance is a native `<script>` in the component. `serializeJsonLd`
already escapes the `<` character to its unicode form, which is the
sanitization that guidance calls for, so only the element changes.

Same build, after: three valid tags per page with `WebSite` in `<head>`, and the
injection queue gone entirely.

* fix(docs): scope the flush-separator rule to a container's first separator

`[data-separator]:not([data-separator] ~ [data-separator])` was meant to keep the
first sidebar group flush against the top padding, but `~` only reaches siblings,
so it also matched the first separator inside every expanded folder. Under
Self-Hosting, "Install" lost its top margin and crowded the "Architecture" link
above it — 25px of gap where "Configure" and "Operate" below it had 40px.

`:first-child` expresses the intent directly. Only the four sidebar roots open
with a separator; every nested folder starts with a page, so the intended case
still goes flush and nothing else changes.

* fix(docs): move the flush-separator rule onto the separator component

Keeps the styling with the component that owns it, per the repo standard, and
lets the global rule be deleted outright rather than corrected — `global.css`
now only loses a rule in this PR. Tailwind's `first:` variant compiles to the
same `:first-child` selector, so behavior is unchanged: the build emits
`.first\:mt-0:first-child{margin-top:0}` and the prerendered HTML carries the
class on the separator.
@waleedlatif1
fix(integrations): close regressions found in the final validation sw…
aeb5624 
…eep (#6764)

* fix(integrations): close regressions found in the final validation sweep

An independent read-only audit of the eight integrations merged to staging
today found defects in every one, most of them side effects of the surgery
those PRs performed on already-shipped code.

Data loss and destructive paths:
- cloudflare: restore the shipped subBlock ids on read filters so existing
  workflows keep their DNS/zone/purge filters. Losing them made
  list_dns_records return the entire zone with success: true, which a
  downstream delete fan-out would then target. The colliding write controls
  are renamed instead, chosen by blast radius.
- cloudflare: refuse an update_ruleset_rule that would tear down the rule it
  edits. PATCH is a replace, so an omitted action_parameters unbound the WAF
  managed ruleset and every override under it.
- cloudflare: split the hidden `enabled` control so a value set while drafting
  can no longer disable a live WAF or rate-limiting rule.
- cloudflare: stop `name` leaking into update_dns_record and renaming a live record.
- okta: stop a blank name overwriting a stored group name via the LLM path.
  The block guard covered only the UI.

Broken on the default path:
- microsoft_ad: update_user sent accountEnabled: "" on its own default, so
  every call left at "No Change" failed. Same tri-state defect already fixed
  for forceChangePasswordNextSignInWithMfa; `visibility` fixed alongside it.
- cloudflare: `domain` is required for self_hosted (the default app type),
  ssh, vnc and rdp; add saas_app/target_criteria and drop dash_sso, which has
  no request variant.

Silent wrong results:
- datadog: list_monitors inherited Create Monitor's tag filter and returned a
  filtered list as if complete.
- servicenow: `fields` carried both a JSON body and a projection on the three
  legacy generic operations. The regression test for this fed already-JSON and
  could not fail; it now feeds a real projection.
- splunk: cancel_search_job reported failure on success by parsing an XML body
  as JSON; readSplunkJson now tolerates it.
- okta: sendEmail === true dropped a string 'true', silently skipping the
  deactivation email.

Security:
- mssql: add writetext/updatetext/readtext to the statement screen. \bupdate\b
  cannot match UPDATETEXT, so both were reachable through the read-only path.
- crowdstrike: chunk repeated-query ids. At the published caps a single request
  built a ~68 KB query string, past typical proxy limits.

Also: splunk count=0 unbounded read, splunk pagination totals, the `nobody`
placeholder that reintroduced the namespace bug by copy-paste, okta cursor and
activate controls split per operation, servicenow sysparm_having syntax and two
required controls no longer pre-seeded with consequential values, datadog block
outputs reconciled with tool outputs, and 16 escaped apostrophes that corrupted
the published Entra docs.

One scope removed from microsoft_ad (User.Read.All). Directory.Read.All and
GroupMember.ReadWrite.All were proposed for removal and verified still required;
a test now asserts they stay.

* fix(integrations): surface corrupt Splunk bodies and partial CrowdStrike deletes

Narrows readSplunkJson's non-JSON tolerance to XML. The dispatching and
job-control endpoints answer in XML, but a body that is neither empty nor XML
was meant to be JSON, so swallowing its parse failure handed get_search_results
an empty envelope and reported a lost result set as a search with zero events.

Annotates a batched CrowdStrike delete that fails partway with the IDs its
earlier batches already removed. Falcon cannot roll those back, so a bare
failure left the caller unable to tell what was gone and a blind retry
re-targeted IDs that no longer existed.

Registers the Cloudflare subblock-ID migration the registry-stability check
requires. The suffixed read-filter IDs never shipped in a release and every
block already materializes the restored IDs, so they are dropped rather than
renamed onto values the collision guard would discard anyway.

* fix(integrations): close the three defects Bugbot found in the sweep

An execute rule sent an explicit empty action_parameters object past the new
guard, because presence was checked rather than emptiness. `{}` is the same
payload Cloudflare's schema default produces, so it unbound the managed ruleset
the guard exists to protect.

Datadog's new List Monitors pagination used a bare `Number()`, so a typo or an
unresolved reference in either advanced field reached Datadog as a literal NaN
— the pattern this same sweep fixed for Entra `top` and the Splunk numerics.

Okta's block still marked the group name required on update, blocking a
description-only update that the tool, its merge helper, and the API all accept.

* fix(integrations): confine the Splunk XML tolerance and the CrowdStrike commit list

Splitting the XML tolerance out of readSplunkJson into readSplunkDispatchJson
puts it only on the three dispatching and job-control tools that need it. The
results path can no longer read any non-JSON body as an empty envelope, so a 2xx
HTML interstitial surfaces instead of reporting a search that matched nothing.
The dispatch reader anchors on the one documented `<response>` root, so an
interstitial fails there too.

A batched delete now records the IDs Falcon echoed in `resources` rather than
the IDs that were requested. A batch can answer 200 while reporting per-ID
failures, and naming those as deleted told the caller to drop still-live
indicators from the retry.

* test(crowdstrike): pin batched partial-delete parity with an unbatched request

A 2xx envelope carrying per-ID errors is a partial success, not a failure —
failedWithoutResources fails the operation only when nothing came back at all.
The batched path already reports it exactly as a single request does, with
deletedIds naming what Falcon confirmed and errors naming what it refused. Pin
that so the contract is not mistaken for a swallowed failure.

* fix(splunk): read the dispatch XML envelope instead of discarding it

A dispatch answering in the documented XML form was replaced with an empty
object, so create_search_job and dispatch_saved_search threw a missing-sid error
after the remote job had already been created — stranding a job the caller could
no longer poll or cancel. The envelope is now projected onto the same `{ sid }`
shape output_mode=json produces, so the search ID survives.

Matching only the opening tag also accepted a body cut off mid-transfer, which
on a cancellation reported a truncated response as a successful cancel. The
pattern now spans the closing tag, so a truncated envelope falls through to
JSON.parse and throws.
@waleedlatif1
waleedlatif1 requested a review from a team as a code owner August 16, 2026 04:40
@vercel

vercel Bot commented Aug 16, 2026 •
edited
Loading

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment
Project Deployment Actions Updated (UTC)
docs Skipped Skipped Aug 16, 2026 11:51am

Request Review

@greptile-apps

greptile-apps Bot commented Aug 16, 2026

Copy link
Copy Markdown
Contributor

Too many files changed for review (879 files, 100 file limit).

@gitguardian

gitguardian Bot commented Aug 16, 2026

Copy link
Copy Markdown

⚠️ GitGuardian has uncovered 1 secret following the scan of your pull request.

Please consider investigating the findings and remediating the incidents. Failure to do so may lead to compromising the associated services or software components.

🔎 Detected hardcoded secret in your pull request
GitGuardian id GitGuardian status Secret Commit Filename
36185505 Triggered Generic Password 257029a apps/sim/tools/microsoft_ad/reset_password.ts View secret
🛠 Guidelines to remediate hardcoded secrets
  1. Understand the implications of revoking this secret by investigating where it is used in your code.
  2. Replace and store your secret safely. Learn here the best practices.
  3. Revoke and rotate this secret.
  4. If possible, rewrite git history. Rewriting git history is not a trivial act. You might completely break other contributing developers' workflow and you risk accidentally deleting legitimate data.

To avoid such incidents in the future consider

🦉 GitGuardian detects secrets in your source code to help developers and security teams secure the modern development process. You are seeing this because you or someone else with access to this repository has authorized GitGuardian to scan your pull request.

@cursor

cursor Bot commented Aug 16, 2026 •
edited
Loading

Copy link
Copy Markdown

PR Summary

Low Risk
Documentation and agent-authoring guidance only; no application runtime, API, or security logic changes in this diff.

Overview
Modernizes integration and tooling agent skills and removes duplicated .claude/commands guides so authoring workflows point at .agents/skills as the single source.

Docs CI gate: Several skills (add-block, add-integration, add-tools, add-trigger, validate-integration) now require bun run docs:check alongside generate-docs and integration-catalog:check. Guidance shifts from reverting “unrelated” generator hunks to committing the full regen output, including catch-up pages left stale by earlier PRs—reverting those hunks is documented as breaking CI.

Migration skill hardening: migrate-application-operation adds a “freeze observable behavior before editing” section: baseline legacy routes as ordered programs, characterization tests for auth order, wire shapes, mutation sequencing, side effects, and explicit warnings that projectAudit/afterSuccess timing and shared wrappers can change behavior even when individual calls are reused.

Command cleanup: Large slash-command files (add-block, add-integration, add-column-type, add-connector, add-enrichment, add-feature-flag, add-hosted-key, add-managed-cli, add-model, add-block-preview) are deleted from .claude/commands; repo rules already reference the skills paths for the same content.

Reviewed by Cursor Bugbot for commit aeb5624. Configure here.

@mzxchandra
fix(forks): stop sync demanding config the source never had (#6766)
ad83796 
* fix(forks): stop sync demanding config the source never had

Fork sync manufactured required re-pick rows the source never asked for, disabling
Sync and suppressing the "Fully mapped" badge. Two distinct causes landed on the
same line in `collectForkDependentReconfigs`.

Cause A - a block-level `required` applied to a nested tool param. The collector
runs one helper over both a block's own subblocks and the params of each tool in a
`tool-input`, reading the raw block config for the nested pass. A Jira Get Issue
tool inside an Agent block therefore inherited Jira's `issueKey` required-condition
even though the tool param is `visibility: 'user-or-llm'` - the agent fills it at
runtime, and `createLLMToolSchema` keeps an empty one in the model's schema
precisely so it can. The tool-row editor already strips `required` for anything not
`user-only`; the fork collector was the only surface that did not.

Cause B - demanding a value the source never had. `projectId` hangs off the
credential anchor, so a Jira block on `write` with a blank project emitted a
required row. The proof it is a collector bug: only members carrying a
`selectorKey` are emitted, so the basic selector gated while its advanced twin
(identical `required`, no `selectorKey`) did not - the same empty config blocked or
not purely on a display preference. That is also why toggling to manual mode
"fixed" it.

Neither fix subsumes the other: an emptiness guard alone leaves a *populated*
`user-or-llm` param gating after a parent swap (`effectiveDependentValue` blanks it
when the parent changed), and the visibility rule alone never touches top-level
subblocks.

Both invariants now meet in one predicate, and the tool-row rule is extracted so
the editor and the sync gate cannot drift. Rows are still emitted either way - only
the gating flag changes - so no stored dependent value is orphaned.

Also fixes the placeholder stutter in the re-pick selector ("Select select issue"),
which composed a verb onto titles that already read as instructions.

Verified end-to-end against a forked workspace: before, 3 of 7 rows were required
(including one with a populated source value); after, only a top-level populated
required field gates, Sync enables, and the badge reads "Fully mapped".

* fix(forks): ask emptiness of the raw dependent value, not the coerced one

Pre-landing review caught a silent un-gating in the new predicate: it asked
`isNonEmptyValue` about `rawSourceValue`, which flattens every non-string to `''`
for the wire contract. A multi-select dependent selector stores an array, so a
populated one reported blank and stopped gating the sync. `isNonEmptyValue`
handles arrays and non-strings deliberately - give it the raw value.

Reachable today via zoho-desk `departmentIds`, the one multi-select dependent
selector with a `selectorKey` + `dependsOn`.

Also from review:
- the canonical id is an alias, so it no longer clobbers a param that owns that
  key as its own `paramId` (first write wins)
- drop a redundant conjunct that implied a third state the code cannot reach
- document the present-but-undefined visibility case, which the resolver's
  `buildToolInputSearchConfig` branch produces routinely
- extract the placeholder-noun transform so a bare "Select" title falls back to
  the whole title instead of rendering "Select " / "No  found", and test it

Tests: non-string and empty-array source values, both verified to fail without
the fix; the basic/advanced parity case now pins both shapes rather than calling
`.every()` on an empty array; the indexer mock is reset per test so the new cases
are not order-dependent.

* fix(forks): make the post-sync collector honor tool param visibility too

Greptile caught a real asymmetry, and corrected a wrong assumption in the first
commit: `needsConfiguration` is NOT warning-only. `promote.ts:1035` skips the
target's redeploy for any workflow in that list, and `:792` also withholds its
chat-deployment carry-over.

So with only the pre-sync collector fixed, an Agent block whose Jira `issueKey`
was populated in the target and cleared by a credential remap would let the sync
through (the modal correctly treats a model-supplied param as non-blocking) and
then silently decline to redeploy that workflow - leaving the fork running its
previous deployed version with no gate and no error.

Both collectors now resolve `required` through one shared
`resolveToolParamRequired`, so the pre-sync gate and the promote path cannot
disagree about what a nested tool param means. The lookup (sub-block id, then
canonical param id, then fail closed to the block-level rule) lives in one place
instead of being duplicated.

The `@/tools/params` mock in remap-references.test.ts is now overridable so a
test can opt into an authoritative resolution; its defaults are unchanged and
the other 74 tests pass untouched. The new case is verified to fail without the
fix.

* style(forks): use TSDoc for the new test declaration comments

CLAUDE.md requires TSDoc for documentation and no non-TSDoc comments. The
vi.mock boundary note, the resolve helper, and the beforeEach mock-reset
rationale all document declarations, so doc tooling could not associate them.
In-body step comments are left as-is - they explain a flow, not a declaration.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

5 participants

@waleedlatif1 @icecrasher321 @TheodoreSpeaks @j15z @mzxchandra