wave 3.5: Effect-native internals #152

Activity

1. ScriptedAlchemy

   added

   enhancementNew feature or requestNew feature or request

   meta-frameworkAgent Bundle compiler-coupled meta-frameworkAgent Bundle compiler-coupled meta-framework

   on Sep 1, 2026
2. ScriptedAlchemy

   added 2 commits that reference this issue on Sep 1, 2026

   refactor(runtime): Effect-native dispatcher and stream pipeline (wave…

   ...

   337d3c7

   refactor(runtime): Effect-native dispatcher and stream pipeline (wave…

   ...

   d691cdb
3. ScriptedAlchemy commented on Sep 1, 2026

   

   ScriptedAlchemy

   on Sep 1, 2026

   OwnerAuthor

   More actions

   Stage 2 (Effect-native dispatcher and stream pipeline) merged in d691cdb (#156).

   Root cause of the final event-count hang, verified by bisect + instrumentation: teardown called cancel() on the Flight ReadableStream from an acquireRelease finalizer while the Flight client held its reader; the locked-stream rejection became an Effect.promise defect during scope close and wedged the event stream's exit. Fixed with scopedAbortSignal + Stream.interruptWhen on the Flight source (no interaction with the locked stream). Details and stage-3 helped/hurt notes in the PR body.

   Parity: runtime 144 pass / 1 skip; example 168 pass / 6 skip; typecheck / lint (boundary-lint green) / eval:spot green; cold start 41.82ms vs 39.74ms baseline (+5.2%, under the 15% gate); event-count test 30/30 stress runs.
4. ScriptedAlchemy commented on Sep 1, 2026

   

   ScriptedAlchemy

   on Sep 1, 2026

   OwnerAuthor

   More actions

   Wave 3.5 is complete. Stage 3 (dev seam) landed one subsystem per PR, strictly parity-gated — every existing dev/workbench suite passed unchanged, zero pin flips across the stage.

   PR	Subsystem	Merge SHA
   #158	Dev-seam boundary module (packages/agent-bundle/src/effect/boundary.ts, exact effect@4.0.0-rc.112 pin, boundary-lint coverage)	70deb9987a0e81f7471764e317c0b963ff6e5116
   #159	MCP session lifecycles as scoped resources; #134 fail-closed contract on the typed error channel (McpSessionStaleEpochError)	0343e04a59740015c6e3cd187c31de22e2674838
   #160	Coalescing rebuild scheduler on fiber + Semaphore(1) + shared Deferred	87711c4f81b2c85bfccf51dd1fc9e41439a93775
   #161	EpochStore staging/leases/recovery: process-wide lease mutexes as Semaphores, publication as an Exit-aware compensation saga, retention via forEach + per-element Exit	f596377618a634b7c0ba76d106886a8da76bd80e
   #162	Conventions doc: stage-3 outcomes, helped/hurt addendum, unstable-adoption record (none)	44e31b56c0ea7debfff0b5fd9bf8764934e80f41

   One honest exception, recorded in docs/effect-conventions.md: ProjectEventHub (the SSE hub) stays imperative. Its public contract is synchronous re-entrant fan-out — foreground shutdown depends on same-tick tombstone delivery (documented in foreground-server.ts) and ~15 tests pin sync replay/removal/ordering — which Effect PubSub (an async fan-out structure) does not model. The SSE wire contract (sequence numbers, replay-gap frames) is untouched either way.

   Parity evidence per PR is in each PR body; the stage-3 tail runs were full unit pool 1999/0 plus the epoch-heavy integration suites 35/0, lint and typecheck green. Stage 4 (policy flip: new wave 4–6 internals born Effect-native behind Promise edges) is now in force.
5. ScriptedAlchemy commented on Sep 1, 2026

   

   ScriptedAlchemy

   on Sep 1, 2026

   OwnerAuthor

   More actions

   Wave 4 Lane A / #97 proof dependency is landed.

   • Thin event-route clients + Effect-native local transport: feat(events): render native hooks through warm runtime #180 (363890b69d475d7a79ef50c4067fb49a9158ca87)
   • examples/rsc-agent-runtime/src/events/tool/after.tsx migration + Claude/Codex schema-conformance replays: feat(examples): adopt semantic event route #182 (9004850090061bab8bacce1ab8635819399b39e8)
   • pnpm eval:spot is green against the compiler-generated Claude event-route wrapper.

   Lane C can proceed from current main.
6. ScriptedAlchemy commented on Sep 1, 2026

   

   ScriptedAlchemy

   on Sep 1, 2026

   OwnerAuthor

   More actions

   Wave 4 Lane C — #103 stage 2 landed: #190, squash 3ae77210e.

   Proof levels shipped, each labeled in provenance and in every failure message:
   mcp-in-memory (real generated MCP server + real MCP client over the SDK's
   in-memory transport pair — protocol contract only, explicitly not host
   evidence), cli-dispatch (argv through the routed CLI's own shell, in-process,
   not a spawned binary), packed-stdio (built artifact's generated entry as a
   separate OS process over real stdio — the only process evidence). Plus
   expectEvents over the #140 render-event stream: sequence-tolerant matching
   with the #120 monotonic lesson baked in as its own matcher.

   Cost rules held: the packed pool gained no new build and no new pack — the
   single journey draws the run-level shared tarball, installs once with the #109
   cached args, builds one artifact, spawns one server, and iterates every
   per-route assertion inside that one session. The new test:projection pool is
   in-process only (22 tests, no build/pack/spawn).

   Coordination outcomes:

   • Lane B (feat(cli): routed CLI from src/cli/** routes (#102 stage 2) #178): reconciled rather than stubbed. invokeCli dispatches
     through the product's own runGeneratedCliEntry over the compiled command
     graph the test manifest now carries (cliCommands), so resolution, argv
     projection, help, --version, and the exit-code policy are the product's.
     The runRscCli fallback was not needed, so there is no follow-up here.
   • Lane A (feat(examples): adopt semantic event route #182): landed in time, so the PostToolUse proof ships in this PR
     rather than as a follow-up. examples/rsc-agent-runtime gains a route-unit
     pool proving the migrated src/events/tool/after.tsx renders from a real
     Claude PostToolUse envelope as a compiled event-route, and the same route
     renders through renderRouteEvents in the projection pool.

   Two defects the packed level surfaced, both fixed here (Effect-native internals
   kept behind Promise edges per this issue's policy, and the harness's public
   surfaces stay Promise-shaped):

   1. Generated executables crashed on any JSX-authored route. Route entries
      were bundled without the React plugin, so JSX lowered to the classic
      React.createElement factory that no generated entry or Flight worker has in
      scope — every documented .tsx route failed at run time with React is not defined. It survived because every build-level test authored routes with an
      explicit createElement import.
   2. Event routes rendered with the wrong props. A route written against the
      public AgentEventRouteProps received undefined for canonical and
      native; renderRoute now performs the same unwrapping the generated Flight
      worker does.

   Core-first note for anyone touching the generated MCP entry: the warm Flight
   host, artifact epoch, route registration, MCP projection, and event-runtime
   lifecycle now live in src/mcp-server-runtime.ts, which the generated entry
   aliases and feeds data. Templating any of that back into the entry would split
   the artifact from what the in-memory level proves; tests/entry-shell.test.ts
   asserts against exactly that.

   ⚠️ Heads-up, unrelated to this lane: packages/agent-bundle/tests/dev-services.test.ts
   ("creates an exact deeply frozen root-independent project context") fails on
   current main with a differing project-context modelDigest. Reproduced on a
   clean origin/main worktree at 6baf597; it passes at #183, so it arrived
   with #184/#185. Whoever owns the skill-IR model digest may want a look.
7. ScriptedAlchemy commented on Sep 1, 2026

   

   ScriptedAlchemy

   on Sep 1, 2026

   OwnerAuthor

   More actions

   dev-services modelDigest regression — fixed in #191

   Root cause: PR #185 (#185) added skillIr, hostDocuments, and skillTreeLayout to NormalizedSkill, but canonicalizeNormalizedModel was never updated to root-relative those new fields. Absolute temp-directory paths (skillIr.source, skillIr.resources[].source, etc.) leaked into modelDigest, so two byte-identical fixtures in different directories produced different digests — exactly what creates an exact deeply frozen root-independent project context asserts against.

   Fix: Extend canonicalizeNormalizedModel to canonicalize Skill IR and host-document paths (source, resource, sidecar, and diagnostic paths). Product-code change with a patch changeset.

   Verification: dev-services.test.ts (26/26), digest-adjacent tests (dev-artifact-service, package-identity, manifest), pnpm typecheck, pnpm lint. Release gates green on #191.

   Note: full Verify matrix still reports unrelated failures on current main (hooks.test.ts, runtime-playground.e2e.test.ts from #186); those are outside this lane.
8. ScriptedAlchemy

   closed this as completedin #191on Sep 1, 2026