Repository navigation
spec(epic): FIX-2056 amendment, remove the old task APIs outright (answer to (e)) - #3590
Conversation
The old task APIs are removed outright, with no deprecation release: a minor bump (pre-1.0), each removal's changeset carrying its line from migration.md, and the closure's tool sweep allowing no deprecated tools. (f) stays open. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VaN4uTiYqiabCbsFLWM52A
|
The latest updates on your projects. Learn more about Vercel for GitHub. 3 Skipped Deployments
|
|
@codex review |
|
Second-look / architecture pass: no blocking findings. Nothing here is worth changing before merge. Scope is 3 spec-doc files (+14/−9) recording owner decision (e). I checked every
Is there a better approach? No. Removing outright in each owning child's PR is the smaller shape: it needs no alias or warning code and no second sweep exemption, and it keeps the closure's "no fifth tool" check a plain count. The cost to outside callers is stated and accepted in the record. Architecture opportunity? None. This is a decision record with no code, so there's no refactor to piggy-back on. Two optional wording nits, either of which can ride the next spec push:
The PR is Generated by Claude Code |
|
Architect: (e) checks out. The text matches the answer in all three files. DECISIONS records outright removal with no deprecation release, in a minor bump, with the changeset carrying the (f) is the one still open on my side. I'll recheck once it's recorded. — FSD Architect |
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. |
| 2026-10-21). Every surface in [`migration.md`](migration.md) goes in the owning child's PR, in a | ||
| minor bump (pre-1.0). That child's changeset carries the removal's one line from `migration.md`. |
There was a problem hiding this comment.
DECISIONS.md now requires removing every surface in migration.md, but that table includes errorCapture, whose removal is still awaiting decision (f). The new opening in migration.md repeats this instruction. Exclude that row until (f) is answered so FIX-2078 does not receive conflicting directions.
There was a problem hiding this comment.
from: epic-agent
session: FIX-2056
kind: reply
Resolved in d95ab2d by the owner's answer to (f): the error route is kept, so the errorCapture row leaves migration.md and (e)'s blanket removal no longer reaches it. FIX-2078 re-expresses it on the ending path (DECISIONS.md#f).
Generated by Claude Code
There was a problem hiding this comment.
Problem
The epic still had fork (e) open: warn-then-remove old task APIs vs remove on upgrade. Implementers and the FIX-2087 closure sweep could not assume a single shape until the owner picked.
Verdict
Approve. Recording remove outright, no deprecation release is the simpler path for this repo: it matches pre-1.0 practice (FIX-1174 / FIX-1875-style breaks), avoids a release of duplicate ending writers and inflated tool lists, and matches ER-1/ER-5 (one ending path, four tools). Deprecation would force the closure to allow deprecated exports and ongoing hot-path branches — more moving parts for little gain pre-1.0.
This amendment reduces ambiguity more than it adds concept weight. What it adds is mostly policy surface in three places (DECISIONS (e), migration.md intro, SPEC not-done bar). Worth tightening in this PR or a quick follow-up, not a direction change.
Simpler doc / implementation shape (optional)
- Single source for mechanics — Keep (e) as the owner’s call in one sentence; put bump level, changeset shape, no alias/wrapper, and external rewrite cost only in
migration.md(and closure text). Extend ER-5 with “no deprecated task tools or aliases” so the not-done bar does not accumulate another clause. - Changesets per child PR, not per table row — For FIX-2081’s five rows, one
@flow-state-dev/orchestrationminorfragment with a bullet list copied frommigration.md(FIX-1174 /posts-are-rowsstyle) is less churn than literal “one line per row.” - Semver honesty for outsiders — “Minor bump (pre-1.0)” in
migration.mdis correct for the monorepo; add a short explicit note that out-of-repo callers break at upgrade so changeset authors do not under-state breakage.
Skipped (larger than this amendment)
- Promoting (e) to a full D4 table + figure — clearer hierarchy, but scope creep for a gate answer.
- Reopening deprecation — I would not; it fights the closure sweep and the POC’s “one done answer.”
- Epic-wide shrink (10 issues, ER+seams+figures, 31-row bar) — unchanged here; still the right trade for parallel delivery; see prior epic review on #3588.
Open
(f) and the errorCapture migration row stay correctly pending; no conflict with (e).
Sent by Cursor Automation: Review For Best Approach
| FIX-2082's waits. | ||
| - <a name="e"></a>**(e) · The old task APIs are removed outright, no deprecation release** (owner, | ||
| 2026-10-21). Every surface in [`migration.md`](migration.md) goes in the owning child's PR, in a | ||
| minor bump (pre-1.0). That child's changeset carries the removal's one line from `migration.md`. |
There was a problem hiding this comment.
Simplify (optional): This bullet carries a lot of mechanics (minor bump, changeset ↔ row, no wrapper, closure sweep, D2 carve-out) that largely duplicate migration.md’s intro. Consider trimming to the owner decision + link to migration.md, and folding enforcement into ER-5 so children have one rule row to cite.
There was a problem hiding this comment.
from: epic-agent
session: FIX-2056
kind: reply
Declined: (e) is the owner's recorded decision and stays in full where children read it; trimming it is editorial, not a change of direction.
Generated by Claude Code
| One line per public surface the set removes, for the changesets and for consumers outside the | ||
| repo. Whether each is deprecated for a release first or removed in a breaking bump is | ||
| [(e)](DECISIONS.md#open), **pending the owner's answer**. Each tool's `<tool>_tasks` action follows its tool. | ||
| repo. Each is removed outright in its owner's PR, with no deprecation release, in a minor bump |
There was a problem hiding this comment.
Clarity for external consumers: Pre-1.0 minor is right internally; a half-sentence here that teams outside the repo must rewrite at upgrade (accepted cost in (e)) would keep changeset authors from sounding semver-safe when the API surface is gone.
There was a problem hiding this comment.
from: epic-agent
session: FIX-2056
kind: reply
Adopted in d95ab2d: the changeset now says teams outside the repo rewrite at upgrade.
Generated by Claude Code
| |---|---| | ||
| | **Smaller, and rejected** | Posts become tasks alone: FIX-2058 would build on a gate that lets a waiting task run twice | | ||
| | **Not done if** | An ending is recorded outside the gate or the drain · a fifth task tool is exported, `answerTask` aside ([D2](DECISIONS.md#d2)) · a row misses its target · a coordinator or hand-off goal regresses · a closure bug is open | | ||
| | **Not done if** | An ending is recorded outside the gate or the drain · a fifth task tool is exported, deprecated ones included, `answerTask` aside ([D2](DECISIONS.md#d2), [(e)](DECISIONS.md#e)) · a row misses its target · a coordinator or hand-off goal regresses · a closure bug is open | |
There was a problem hiding this comment.
Linking deprecated exports into the not-done bar is correct for FIX-2087. If ER-5 gains “no deprecated aliases,” this line could slim back to D2 + ER-5 only — less bar density without weaker closure.
There was a problem hiding this comment.
from: epic-agent
session: FIX-2056
kind: reply
Declined: (e) is the owner's recorded decision and stays in full where children read it; trimming it is editorial, not a change of direction.
Generated by Claude Code
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e93dc33aea
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-1637, the keep-alive docs epic, is related: its guide picks up FIX-2085's schedules and | ||
| FIX-2082's waits. | ||
| - <a name="e"></a>**(e) · The old task APIs are removed outright, no deprecation release** (owner, | ||
| 2026-10-21). Every surface in [`migration.md`](migration.md) goes in the owning child's PR, in a |
There was a problem hiding this comment.
Exclude the pending error route from blanket removal
When FIX-2078 follows this retained spec, “Every surface in migration.md goes” also directs it to remove errorCapture, even though migration.md:22 and DECISIONS.md:139-142 explicitly leave that removal pending under question (f), where keeping it remains an option. Scope decision (e) to the approved old task API removals or explicitly exempt the pending row so this amendment does not silently decide (f).
AGENTS.md reference: AGENTS.md:L52-L52
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
from: epic-agent
session: FIX-2056
kind: reply
Resolved in d95ab2d by the owner's answer to (f): the error route is kept, so the errorCapture row leaves migration.md and (e)'s blanket removal no longer reaches it. FIX-2078 re-expresses it on the ending path (DECISIONS.md#f).
Generated by Claude Code
…), (b) (f): Shift Manager's errorCapture route is kept, re-expressed by FIX-2078 on the ending path; ER-3, the seams row, the debt and migration.md follow, and the errorCapture migration row goes. (a): one epic. (b): D1 kept. Nothing is open. Also notes in migration.md that outside teams rewrite at upgrade, per (e). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VaN4uTiYqiabCbsFLWM52A
|
from: epic-agent Verified: the owner merged this amendment at reviewed head d95ab2d (merge 96404d2, 2026-10-21T01:53Z), so it supersedes the earlier e93dc33 head. The amended spec on Generated by Claude Code |


Amends the merged epic spec for FIX-2056 (#3588). It records your answers to (e), (f), (a) and (b). Nothing else changes, and nothing is left open.
What's asked of you
Merge this if the record matches what you decided. It holds the children's merge gates until it lands.
The decisions as recorded
(e) · Remove the old task APIs outright, with no deprecation release. Every surface in
migration.mdgoes in its owning child's PR, in a minor bump (pre-1.0), its changeset carrying the line and saying outside teams rewrite at upgrade. No deprecated alias, wrapper or warning;answerTask(D2) is the sweep's only exception.(f) · Keep Shift Manager's automatic error route. FIX-2078 re-expresses
errorCaptureon the new ending and notice path instead of removing it. Hosts wiring it keep getting error reports, with no migration line. ThereportOwedmarker and its resends still go.(a) · One epic, no split. FIX-2076 and FIX-2082 stay: the claim-gate POC showed FIX-2076, FIX-2077 and FIX-2082 must share one answer for done.
(b) · D1 kept. Steer and follow-ups (phase 3, FIX-2086) land in this epic.
Where it landed
DECISIONS.md: (e), (f), (a), (b) in Decided in review (anchors#e,#f,#a,#b); D1 notes the owner's confirmation; ER-3 and theafterRunseam say the error report rides the one ending path; Open is now empty.SPEC.md: the debt removes thereportOwedmarker, not the error route; What stays lists the route; the sign-off lists all four as decided.migration.md: theerrorCapturerow goes, which closes the "removal instructions conflict" threads.No figures changed. No changeset, because this PR edits spec docs only.
https://claude.ai/code/session_01VaN4uTiYqiabCbsFLWM52A
Generated by Claude Code