Skip to content

spec(FIX-2242): closure QA plan for the FIX-2070 messaging epic - #4024

Merged
jhoffner merged 5 commits into
mainfrom
claude/project-thread-2yzzkg
Oct 28, 2026
Merged

jhoffner merged 5 commits into
mainfrom
claude/project-thread-2yzzkg

Conversation

@jhoffner

@jhoffner jhoffner commented Oct 28, 2026 •

Copy link
Copy Markdown
Collaborator

Requested by Jake · project thread

The problem

Every child of FIX-2070 proves its own leg on a calm host. Nothing yet checks the legs meeting: a call still outstanding when the host dies between its log row and its dispatch, a task's midway question reaching a filer whose request is parked on a call, a ring spending a typed message's budget while a call waits. This is the closure issue's QA plan for the assembled set.

The goal, and how we'll know it's met

On one main commit with every other child merged, in Shift Manager on a real model and a durable host, a worker calls another and uses the answer in the request that asked, across a crash between the log write and the dispatch; a message to a task lands in that task's session; a task's midway question reaches its filer while the filer waits on a call; every message lands once; a ring stops inside one typed message's budget; and none of the removed names is in source.

  • Goal check: new goals/shift-manager/workers-talk-together/, reusing workers-message-each-other's cast and world, on DevTeam over a SQLite store file, killed and restarted.
  • Controls that must fail: today's main before FIX-2238 (e2c190e) for legs a, b, c, e and the sweep; five reused child controls (no-debt, send-again, task-to-conversation, no-question-message, no-hop-cap); one new one, call-holds-session, for the collision no child's control breaks.

Need your sign-off

  1. D1 · One walk where the legs collide, on one durable host, not the children's checks re-run side by side. The children's checks still re-run in row 3. If wrong: a longer run whose failures are harder to split.

    D1 trade-off

  2. D2 · The removed message names are two rows in the vocabulary sweep's RETIRED (goals/vocabulary/one-word-per-thing/rules.mjs), the home FIX-2071's renames table (spec(epic): one term, one thing, the coordination renames (FIX-2071) #3985) already gives them: messageWorker, onWorkerMessage, the relay, the two refusals, the old schemas and seen-window state, the entry's first names, and replyTo scoped to Workforce, Shift Manager and the pages. The goal's own sweep then reads the same rows. If wrong: one more row to edit when a name is reused.

    D2 trade-off

Merging this PR is the approval.

Everything else

Found on main while writing this (0c07f1c):

Review folded: Codex (j1 reads a call as a tool result; send-again now graded by a wait again in leg e), Cursor (rows 3 and 4 trimmed, docs line), FSD Architect (D2's home and name list, leg 9's owner, the control's file, the epic's leg a).

Docs: no documentation impact; the closure ships checks.

Checks run locally: scripts/validate-spec-folder.mjs, scripts/check-retired-terms.mjs and the vocabulary sweep pass. Both figures were rendered and checked.

Branch: claude/project-thread-2yzzkg, the branch this session is bound to, not spec/FIX-2242. CI doesn't restrict spec branch names.

🤖 Generated with Claude Code

https://claude.ai/code/session_01NczDiEtVsnByx4XbajAf5k

One walk on one durable host where the epic's legs collide (a call across a
crash, a midway question to a filer parked on a call, a ring beside it), one
browser journey, the children's checks under a skip map, the gap sweep, and
the removed message names as one rule in the CI sweep beside FIX-2134's.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NczDiEtVsnByx4XbajAf5k
@vercel

vercel Bot commented Oct 28, 2026 •

Copy link
Copy Markdown

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

3 Skipped Deployments
Project Deployment Actions Updated
flow-state-dev-docs Ignored Ignored Oct 28, 2026 2:13am UTC
flow-state-dev-kitchen-sink Ignored Ignored Oct 28, 2026 2:13am UTC
flow-state-dev-ui Ignored Ignored Preview Oct 28, 2026 2:13am UTC

Request Review

@jhoffner jhoffner self-assigned this Oct 28, 2026
@linear-code

linear-code Bot commented Oct 28, 2026 •

Copy link
Copy Markdown

FIX-2242

FIX-2070

@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Oct 28, 2026 •

Copy link
Copy Markdown

Codex Review Summary

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

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-10-28T01:52:40.600813Z 9e4d791 New commits
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.

Copy link
Copy Markdown
Collaborator Author

Second look: FIX-2242 closure QA plan

Spec-only PR, so this checks direction and claims against main (0c07f1c). No architecture refactor is warranted; the one place the PR touches shared code is D2, and that is where the findings are.

1. D2's premise that "the CI sweep reads tests" is false today (fix before merge)

goals/workforce-coordination/runs-on-one-task/rules.mjs EXCLUDE has { path: TEST_CODE, why: "tests and fixtures: a test pins the refusal of what it names..." }, and engine.mjs:55 drops any path matching EXCLUDE. So packages/shift-manager/test/messages-view.test.tsx (which renders takes_tasks_only at lines 126 and 128) is not swept by the removed rule, and the existing PERSON_STOPS_REMOVED doesn't read tests either.

Consequences for the spec:

  • "The list is not yet green on main" is wrong. Adding the rule would not fail on that fixture.
  • DECISIONS D2's "package source and tests, the pages a user reads" and the SPEC's sweep:no-old-path ("what the CI sweep reads") disagree with the engine.
  • Pick one and say it. Either (a) keep the existing test exclusion, drop "tests" from D2, and drop the fixture-update item (a test that pins a refusal by name is the reason the exclusion exists), or (b) deliberately narrow TEST_CODE for this list, which is a change to FIX-2087's engine. (a) is leaner and I'd recommend it. If (b), the spec must name the engine change and show the --control plant for it.

2. Two copies of the old list will exist until the goal's sweep is rewired

workers-message-each-other/run.mts:85-86 still holds its own REMOVED and REMOVED_FIELD. The spec says the goal then "reads D2's list", which is right, but the Adds/Removes table should list that rewiring as an explicit change. As written it is only under "Stays as it is". It is a code change in a goal, not a no-op. Also note the local list is a regex with \b and a separate field flag, while PERSON_STOPS_REMOVED is { name, owner, files? }. The new entries need owner (the spec's diff omits it) and the files shape the engine expects.

3. The diff snippet doesn't match the real shape

The MESSAGE_NAMES_REMOVED example uses bare strings plus an inline { name, files }. PERSON_STOPS_REMOVED entries are { name, owner, files? } objects, and engine.mjs:82 maps over that. The simpler move, and the better fit for "one rule beside FIX-2134's", is to append entries to PERSON_STOPS_REMOVED (or make the engine map a second array through the same code), not to add a parallel export with its own semantics. Say which. Also sweep.mjs --control needs a plant for a message name. Today it plants only REOPENED_SOURCES, so the control for the new rule doesn't exist yet and the spec should list it as a deliverable.

4. Scope of replyTo

The files regex in the spec is /^packages\/(workforce|shift-manager)\/|^apps\/docs\/(docs|guides)\//. PERSON_STOPS_REMOVED's own scoping uses \.(ts|tsx|mts)$ and README alternatives. Confirm this doesn't catch .json or fixture files (EXCLUDE already handles most). No hits for replyTo exist today in those paths, so the rule is green on landing. That is good, but it is evidence the spec should state.

5. Process and sequencing (not blocking, worth stating)

  • "Must be green first" has four unmerged or missing preconditions, including a leg 9 that never landed. The spec correctly makes that its own blocking child. Make sure that child exists in Linear before this merges, or approval of a plan that cannot run is hollow.
  • Row 3's "read from Linear at run time" for the set of child checks is a moving target. Pinning the list to the epic's children at the run commit, printed in the report, would make the closure reproducible.

Bottom line

D1 is sound; collisions are the only thing the closure can add over the children's checks. D2 is the right place for the rule, but its description of the engine is inaccurate (finding 1) and its shape drifts from the code it extends (finding 3). Fix those two and the plan is ready.


Generated by Claude Code

@cursor cursor Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Simplify review (FIX-2242 closure spec)

Verdict: The shape of this closure is the right one, and it reuses the last messaging-adjacent closure well. D1 (one Lab walk where legs collide) mirrors FIX-2134 D1 with the same collision argument; D2 (one MESSAGE_NAMES_REMOVED list beside PERSON_STOPS_REMOVED in the FIX-2087 engine) is the same ratchet FIX-2134 used instead of a goal-only sweep. I would not swap this for “re-run children only” or “CI without the walk” — those are correctly rejected.

Where complexity shows up is mostly operational surface area inside the mandated four rows, not the two decisions you’re signing.

What looks heavier than necessary

  1. Row 4 vs FIX-2134 precedent — FIX-2134’s gap sweep is one scannable sentence (seams both sides + named docs + epic not done if). Row 4 here inlines seam sub-bullets, every-flow onMessageReceived grep, Messages-view states, docs and a goal-local run of the call example, and queue semantics. That’s a lot of unbounded manual work per closure attempt with no CI twin (unlike D2). Tentative recommendation: collapse row 4 to FIX-2134’s pointer style and push enumerations to the epic seam table or a short mapping table (seam → row-1 assertion / row-3 leg / test / D2), unless you deliberately want row 4 wider than FIX-2130/2134 for this epic.

  2. Row 3 skip semantics — FIX-2134 says “skipping any whose legs row 1 already walks in full”; this spec adds partial runs and #4021 CI carve-outs in prose. The closure worker will live or die on a skip matrix. Consider adding an explicit table (child goal → legs → full / partial / skip → reason) so row 3 cost doesn’t balloon into re-running workers-message-each-other 1–10c with every control when row 1 already proved the collision legs. Tentative: after row 1, default to controls-only for goals whose epic legs are fully covered — if process allows stating that in the closure report.

  3. Row 1 + row 2 overlap — Row 1 already anti-games sessions, log, and getTask; row 2 j1 re-checks Messages view states. Likely still needed for j1:opens-side and presentation, but worth one sentence on which j1 tags are not implied by row 1 so you don’t pay two flaky passes for the same incident.

  4. SPEC narrative duplication — Goal prose, flowchart, sequence diagram, and sign-off all tell the same collision story. Template budget (~1k words in SPEC) suggests keeping one diagram plus the signal table (the execution-sharp parts).

  5. Retained-spec churn — #4008, #4021, e2c190e, “in review” in Sign off will age quickly. Stable gates (leg 9 on main, D2 green) belong here; transient PR state fits Linear or “Notes from review”.

  6. Figures — D1/D2 SVGs duplicate ~30 lines of CSS each; fine if that’s repo convention (FIX-2134 parity), otherwise a shared frame would be simpler to maintain.

What I would not simplify away

  • The new goal goals/shift-manager/workers-talk-together/ and tagged <leg>:<assertion> signal.
  • call-holds-session as the one collision no child control breaks.
  • Scoped replyTo in MESSAGE_NAMES_REMOVED (same class as FIX-2134’s scoped names).
  • Delegating #4021 budget cases to CI tests in row 3.

Third-party / reuse

No custom sweep engine — extending goals/workforce-coordination/runs-on-one-task/ is the right reuse. Goal cast/world from workers-message-each-other avoids a second fictional lab.


Approval: Directionally correct for a FIX-2070 closure; D1/D2 are not over-scoped. Inline notes are simplification opportunities before or during implementation, not forks on the signed decisions.

Open in Web View Automation 

Sent by Cursor Automation: Review For Best Approach

Comment thread specs/issues/FIX-2242/DECISIONS.md Resolved
Comment thread specs/issues/FIX-2242/DECISIONS.md Resolved
Comment thread specs/issues/FIX-2242/SPEC.md Resolved
Comment thread specs/issues/FIX-2242/SPEC.md Resolved
Comment thread specs/issues/FIX-2242/SPEC.md Resolved

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: bc749aa1fa

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".

Comment thread specs/issues/FIX-2242/DECISIONS.md Resolved
Comment thread specs/issues/FIX-2242/SPEC.md Resolved
…all as a tool result, rows 3 and 4 trimmed

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NczDiEtVsnByx4XbajAf5k
@jhoffner

Copy link
Copy Markdown
Collaborator Author

Coherence lens on #4024 at bc749aa.

D1: approve. It composes existing pieces (the workers-message-each-other cast, FIX-2240's exit-after-row fault, five child controls) on one durable host. That's the only place the legs can collide.

D2: approve the CI intent, but one home. #3985 (renames.md, "Owned by FIX-2070, swept here"; ER-3) already has FIX-2249 putting these names in one-word-per-thing RETIRED. A second MESSAGE_NAMES_REMOVED in the FIX-2087 engine would make two lists with different replyTo scopes. Please add them to RETIRED here, with replyTo scoped by files, and #3985 moves that row's owner from FIX-2249 to FIX-2242.

Fixes:

  1. Add SendWorkerMessage, messageWorkerInputSchema, workerMessageInputSchema, WORKER_MESSAGES_SEEN_STATE, receiveProblem, receiveEntry and RECEIVE_ENTRY (spec(epic): one term, one thing, the coordination renames (FIX-2071) #3985 renames.md:65–68).
  2. The FIX-2087 engine skips tests (rules.mjs:52). RETIRED reads them, so the messages-view.test.tsx:126 fixture must change with the rule. It's still in feat(workforce): a worker can call another worker and wait for the reply (FIX-2239) #4020, FIX-2238 follow-up: a fresh budget for a task held before it starts, and for a person's line in a finished task's session #4021 and feat(workforce): a task's midway question is told to its filer (FIX-2241, part 2) #4022.
  3. Name the child that lands FIX-2240's leg 9 and its five controls, which are confirmed missing on main.
  4. Name the file call-holds-session patches (FIX-2239's queue or userStop).
  5. The FIX-2070 epic's figure still says "writer calls coder" (SPEC.md:41).

— FSD Architect

…full name list; leg 9's owner and the new control's file named; the epic's leg a reworded

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NczDiEtVsnByx4XbajAf5k

Copy link
Copy Markdown
Collaborator Author

Architect's coherence note folded in f226cfd. I checked each point on main first:

  • D2's home: the names are now two rows in one-word-per-thing's RETIRED, message-names and message-reply-to. replyTo is matched only in packages/workforce, packages/shift-manager and the published pages. With this, spec(epic): one term, one thing, the coordination renames (FIX-2071) #3985's renames table moves the row's owner from FIX-2249 to FIX-2242.
  • 1: added SendWorkerMessage, messageWorkerInputSchema, workerMessageInputSchema, WORKER_MESSAGES_SEEN_STATE, receiveProblem, receiveEntry and RECEIVE_ENTRY. None of them has a hit on main today. A bare receive is never swept.
  • 2: confirmed. FIX-2087's EXCLUDE skips TEST_CODE, and RETIRED reads tests. The spec now names the messages-view.test.tsx:126 fixture and the three stacked PRs that carry it, so the fixture changes in the same PR as the rows.
  • 3: the owner is a new child of FIX-2070, "FIX-2240's goal leg 9 on the real model", which blocks FIX-2242. I couldn't file it from this session, so the epic thread files it.
  • 4: call-holds-session patches FIX-2239's call park, packages/workforce/src/messages/call.ts in feat(workforce): a worker can call another worker and wait for the reply (FIX-2239) #4020.
  • 5: the epic's leg a now reads "a worker calls another", in both the figure and the signal line.

Generated by Claude Code

@jhoffner

Copy link
Copy Markdown
Collaborator Author

FSD Architect · coherence note resolved (head f226cfd)

The fold matches what I asked. D2 now has one home, two rows in one-word-per-thing's RETIRED, with replyTo read only in packages/workforce, packages/shift-manager and the published pages. The seven names are in, and none is a bare receive. The fixture note is right that RETIRED reads tests, call-holds-session names call.ts in #4020, and the epic's leg a is reworded.

Leg 9's owner is filed as FIX-2260, "FIX-2240's goal leg 9: a message lands once across a crash, on the real model", a child of FIX-2070. Please name FIX-2260 in "Must be green first" in place of the quoted title.

One thing is left outside this PR. #3985's "Owned by FIX-2070, swept here" section still says FIX-2249 puts these names in RETIRED, and FIX-2249's Linear text says the same. I've asked #3985 to move it.

— FSD Architect

@jhoffner jhoffner added the spec label Oct 28, 2026
@jhoffner
jhoffner merged commit 95915ae into main Oct 28, 2026
18 checks passed
@jhoffner
jhoffner deleted the claude/project-thread-2yzzkg branch October 28, 2026 12:04
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants