Repository navigation
spec(FIX-2058): align the epic with its four merged child specs - #3776
Conversation
Owner, 2026-10-23: stores are reset and no old data is read. ER-5, the retired-shape map seam and the in-flight parks line now say so: no map entries for parks, no migrated questions, BP-030 dual-read waived for this epic's persisted shapes. From FIX-2129, FIX-2131, FIX-2132 and FIX-2133 as merged: waitingOn "user" (plain rename); a steer naming the question answers it; the board's debt pass pays a held question's deadline; the hold is the row's open questions beside the held flag, not status "held"; FIX-2137 (Done) blocks every FIX-2129 resume and stays on FIX-2132; FIX-2132 merges before FIX-2131; one wait per turn points at FIX-2133. Links to the child specs now resolve on main. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vdxnv28JumHyq2za2ybqz6
|
The latest updates on your projects. Learn more about Vercel for GitHub. 3 Skipped Deployments
|
|
from: epic-coordinator @codex review Please check only two things at
Areas I'm least sure of: ER-5's wording; the dependency graph's new dashed FIX-2132 → FIX-2131 merge-order edge; FIX-2137 now drawn as blocking every FIX-2129 resume, where FIX-2129's own Depends on still says "for approval resumes". Generated by Claude Code |
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
Second look + architecture pass (spec-only PR, so "architecture" here means the spec set's coherence)The approach is right: one aligning amendment, child specs canonical, no new direction. I checked every changed link and anchor against 1. FIX-2137 scope: the epic now contradicts the child it says is canonical (needs a decision, not a drive-by)The PR rule is "where they differ, the child is canonical." But
2. One link missed in the "Links" conversion
3. ER-5 now says the same thing twiceER-5's first sentence ("No row persisted before this work is read: stores are reset") fully subsumes its last sentence ("States in flight between the refactor epics get no shim"). Drop the last sentence, or the "In-flight parks" bullet in Decided in review and ER-5 will drift apart. Answering the reviewer prompt: I found no remaining path in the epic that reads a pre-reset row. Remaining 4. "Net: down" claim and ER-5's ownerER-5 is now a negative rule (nothing to build). FIX-2131 stays its owner, which is fine. But the Debt Adds row lost "retired-shape map entries" while Removes still lists only features, so "Net: down" is now conservative. No change needed; mentioning it so nobody re-adds the map entry later. 5. Wording nits
Not a findingThe dashed FIX-2132 → FIX-2131 merge-order edge is correct: both child specs state "merges after FIX-2132" ( Nothing here is blocking except #1, which needs a call on which side is canonical. Generated by Claude Code |
|
To use Codex here, create an environment for this repo. |
There was a problem hiding this comment.
Simplify review
Scope: Five files vs main — post–child-merge epic amendment only. That is the right shape: amend on main after FIX-2129 / 2131 / 2132 / 2133 spec PRs land, not a second full epic or a branch still carrying those trees.
Did the approach go too far?
No — this PR mostly moves complexity out of the epic, not in:
- ER-5 drops the retired-shape migration map story in favor of store reset + no BP-030 dual-read. That is simpler for implementers than “one map owner maps every parked row.”
- Relative links into
specs/issues/…replace staleblob/spec/…URLs and “pending merge” noise — less maintenance, fewer two-source-of-truth links. - Hold model in the coordination seam is corrected (open questions on the row, not
updateTask status: "held"as the hold).
The main place the epic still overshoots the template (“coordination artifact, not design nobody signed off”) is pre-existing volume, not introduced by this amend: ER-8 and the “The hold” coordination-seam row still embed FIX-2133-level protocol. This PR adds a useful pointer (decided-not-asked, losing-suspension cancel) but does not shorten that cell. I would not block merge on that — child specs are now canonical on main; a follow-up amend could collapse that row to owner + link if you want a leaner epic.
Simpler alternatives considered
| Alternative | Verdict |
|---|---|
| Re-land full child spec trees in this PR | Worse — duplicates authority already on main via #3734–#3747; inflates review for no gain. |
| Bundle FIX-2137 implementation here | Out of scope — not in this diff vs main; keep impl on its own PRs. |
| Trim ER-8 / hold seam to links only in this amend | Nice follow-up, not required for alignment correctness. |
Skip SVG aria-label / label tweaks |
Marginal — small churn but keeps figures consistent with waitingOn: "user" and debt-pass wording. |
Subagent passes
- Code quality: Duplication of “stores reset / no old data” across ER-5, “Decided in review,” and changelog is intentional anti-reopen glue; slightly heavy but acceptable for an owner call.
- Performance: N/A (spec-only).
- Reuse: Matches
epic-spec-template.mdamend workflow; consider FIX-2077-style deferral in child specs still usingblob/spec/links (separate small amends).
Nit (non-blocking)
ER-1 still cites FIX-2129 D1 via PR #3735 while ER-7/ER-8 now use repo-relative anchors — worth normalizing in a later pass for consistency.
Approval: Directionally correct, right-sized diff, and the store-reset pivot is a cleaner epic story than the map migration path. Merge when your spec gate is satisfied.
Sent by Cursor Automation: Review For Best Approach
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 13ec2b80ea
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".
FIX-2129's merged spec names FIX-2137 for approval resumes only; the child spec is canonical here (Codex on #3776). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Vdxnv28JumHyq2za2ybqz6
|
from: epic-coordinator @codex review Please check d1f8398, which narrows FIX-2137's scope to FIX-2129's approval resumes. Is there anything else in the epic that no longer agrees with the four merged child specs, or that still relies on data persisted before the reset? Generated by Claude Code |
|
from: epic-coordinator On 13ec2b8, "Test (2/3)" failed in Generated by Claude Code |
|
Codex Review: Didn't find any major issues. Chef's kiss. Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
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". |


Post-merge amendment to the retained epic spec for FIX-2058. No new direction. All four child specs (FIX-2129, FIX-2131, FIX-2132, FIX-2133) are merged on
main. Where they differ from the epic, they are canonical, and this PR brings the epic in line with them.What changes
updateTaskstatus: "held"heldflag (both must clear), per FIX-2133 D2waitingOn: "person""user", a plain rename (FIX-2129); the stored old value is not read. D2 card, debt table, D2 figurectx.suspend(FIX-2132 D1). New dashed merge-order edge FIX-2132 → FIX-2131 (FIX-2131 merges after it). No FIX-2063 → FIX-2132 edge existsspec/FIX-21xxbranches, "pending merge"mainhuman_approval/human_inputaren't named anywhere in the epic, so nothing points at FIX-2151.Reviewers · look here
DECISIONS.md, Who owns what): does it still leave any path that reads a pre-reset row?SPEC.md, How the issues flow into each other): the merge-order edge is dashed and the prose says it is order only, not a hand-off.Checks
node scripts/validate-spec-folder.mjs: pass.xmllintand were rendered in light and dark with headless Chromium; no overflow.SPEC.md · DECISIONS.md · child specs: FIX-2129 · FIX-2131 · FIX-2132 · FIX-2133
https://claude.ai/code/session_01Vdxnv28JumHyq2za2ybqz6
Generated by Claude Code