Simplify tool and event authoring while preserving JSX

[ScriptedAlchemy]

This replaces overlapping authoring APIs with defineTool and one conventional event system. Events keep JSX: .ts handlers return decisions, .tsx handlers render directly, and ctx.render('./name.view.js', data) loads an explicit sibling view. Providers resolve lazily with the complete request context and trace observations; process identity is exposed as context.process.

Inspection now reads literal metadata instead of reconstructing application behavior. config.inputJsonSchema supplies forms and named CLI flags; commands without it accept --input JSON. Compiled executables validate through the original schema, preserving defaults and transforms. Handlers and mapInput accept sync or async functions. Runtime compiler imports are rejected by the bundler, including aliases, while type-only imports remain legal.

Removes the before/preflight closure extractor, Zod source interpreter, shared alias resolver, routine AB4834 TypeScript Program, semantic import walker, deep function-shape scanner, and obsolete runtime application/operation APIs. Updates both documentation locales, examples, and the single breaking changeset; manifest version 6 records event handlers and CLI JSON mode. The optional source-only unit-render loader remains because its isolation is distinct from normal Workbench execution through generated artifacts. Installation-script dependency policy remains outside this change.

Review follow-ups are fixed: lightweight events mount state/notices and native identities with the user-data state anchor; one deadline covers provider initialization, the handler, deferred child startup, and rendering; native receipts carry resolved lineage plus bounded aggregate provider count/span evidence, while per-provider observations remain in Workbench invocation results. Direct JSX events no longer return the obsolete execute marker, so Workbench renders their documents.

Validation:

• pnpm build, pnpm typecheck, and pnpm lint pass.
• Unit: 4,411 passed / 6 skipped; route-unit: 91 passed; projection: 197 passed.
• Full integration pool: 1,175 passed / 4 skipped.
• Full packed pool: 44 passed / 1 skipped.
• Focused deadline, provider-trace, receipt-limit, privacy, and lineage regressions: 39 passed.
• pnpm docs:site:build passes: locale parity and 0 broken links across 35,933 internal links.
• Independent non-Grok review found no merge blocker after follow-up.

[changeset-bot]

🦋 Changeset detected

Latest commit: 2cb4930

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 3 packages

Name	Type
agent-bundle	Minor
@agent-bundle/runtime	Minor
create-agent-bundle	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-11T02:28:34.821131Z	964dcf9	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[pkg-pr-new]

npm i https://pkg.pr.new/ScriptedAlchemy/agent-bundle@782

npm i https://pkg.pr.new/ScriptedAlchemy/agent-bundle/create-agent-bundle@782

npm i https://pkg.pr.new/ScriptedAlchemy/agent-bundle/rsc-markdown-stream@782

npm i https://pkg.pr.new/ScriptedAlchemy/agent-bundle/@agent-bundle/runtime@782

commit: 97be7ce

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 964dcf957f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[ScriptedAlchemy]

Disposition of the three confirmed findings (fixed in 2cb4930037):

1. P1 deadline handoff — the generated lightweight wrapper now passes the combined AbortSignal.any([controller.signal, AbortSignal.timeout(timeoutMs)]) signal to runExecutor, so provider initialization, handler execution, deferred child startup, and rendering consume one budget. RED: a 1,000 ms route with a 350 ms handler and 1,200 ms view completed late with exit 0. GREEN: it aborts with exit 1 within the asserted 700–1,800 ms window, emits no late output/marker, and the recorded child PID is no longer alive after source deletion.
2. P2 provider timing / receipt limit — overlapping starts are reference-counted into one aggregate phase. The deterministic A(0) / B(10) / B-finish(20) / A-finish(30) case now emits one finish with count: 2 and durationMs: 30 (previously A reported 20). Native receipts compact provider observations to first-start/last-finish plus total count; 16 concurrent providers post and decode successfully, and a 16-provider sequential unit case compacts 34 raw events to four receipt events. Receipts contain no provider value, key, or filesystem path.
3. P2 native lineage — the normal generated lightweight wrapper records native identity and the already-resolved lineage before running the gate, so both completed and failed receipts carry the same lineage visible to the handler instead of not-provided.

Qualification (fresh): pnpm build, pnpm typecheck, pnpm lint; unit 4,411 pass / 6 skip; route-unit 91 pass; projection 197 pass; full integration 1,175 pass / 4 skip; packed 44 pass / 1 skip; docs site 0 broken links across 35,933 internal links. Independent Fable review and follow-up found no merge blocker. The prior PR-body claim is narrowed: native receipts carry aggregate provider count/span evidence; per-provider observations stay in Workbench invocation results.