Repository navigation
spec(FIX-1797): Closure: two users each run a private roster and share a project through workstreams they own - #2835
Conversation
…ster each, one shared project The epic's three legs in Shift Manager with two users and a real model, each change made through the app; a milestone run of leg c on FIX-1788's merge commit; older goal checks the epic breaks held to rewrite-or-retire by the child that broke them; the app-builder journey from the published docs alone. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV
|
The latest updates on your projects. Learn more about Vercel for GitHub. 3 Skipped Deployments
|
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 (direction only)Read SPEC, PLAN, DECISIONS, and checked the repo. The goal and the legs hold up. The claims I could verify do too: 1. D1 depends on a manual step nothing checks (above the bar)The spec admits "no workflow automates it" and "it opens no PR, so the blocked-by relations don't hold it". If the coordinator misses the wake that sees FIX-1788's last PR merge, FIX-1791 and FIX-1795 can merge on an unchecked privacy fix. Nothing fails or reports that. The "hook later" follow-up doesn't help the first epic. Leaner shape: make the milestone a small child of FIX-1786 with its own PR, "leg c harness + milestone report". It is 2. Every finding reruns the whole plan (above the bar)QR-19 reruns all of parts 1 to 4 on a fresh commit, which includes P3.9's ~30 goal runs and real-model turns, after each fix. A finding can come from any part, and the final run needs to be on one commit. But that doesn't mean every fix needs the full cost. Consider: a finding is retested by its own step plus the controls it names. The full plan runs only once the open-findings list is empty, and that run is the one that opens the PR. This keeps "all final checks on one commit" and cuts the repeated model spend. Spec it explicitly, or the implementer will rerun everything by default. 3. S2 misdescribes what exists (architecture note, above the bar)
Notes (recorded for the implementer, not blocking)
Generated by Claude Code |
There was a problem hiding this comment.
Simplify review (FIX-1797 spec + branch contents)
Read the retained spec set and the engine diff that landed on this branch (FIX-1793 P1 / ownerWrites via merge). Parallel read-only passes covered spec orchestration, engine hot paths, and reuse against FIX-1650 closure + goals/lib.
Verdict on approach
The integrated closure (legs a–c in Shift Manager with two users, milestone on FIX-1788, controls on scratch patches) is the right delta over “re-run each child’s goal on one commit.” D1–D3 are well argued; HTTP-only or child-only regression would not meet ER-28/ER-30. I would not shrink the epic proof to integration-tests or Playwright in kitchen-sink — that conflicts with D2 and the anti-game rules you already locked.
ownerWrites as a second mode on the existing owner-key fence (same module, startup fence, registry write hook) is also the cleaner option vs app-only guards or a third-party ACL library — the FIX-1793 POC G1/G2 story still applies.
Where the plan adds complexity beyond what the epic forces, I’d consider trimming before implementation (not necessarily blocking spec direction):
Higher-impact simplifications (spec / process)
-
cad4e2780pinned in five places — This SHA is already stale relative to branch HEAD (P1 merged). FIX-1720-style naming (“commit before first child merged”) with resolve at run time (merge-base / documented first-child merge) would keep c7, the “today’s main” control, part-4git diff, and FIX-1796 census aligned without spec edits on every merge tomain. -
P3.9 runtime import-walk — D3 is real (retire/repair checks the epic broke), but discovering directories at run time is the highest flake and wall-clock cost (many subprocesses × build/boot × some real models). Consider a checked-in manifest (FIX-1650 / FIX-1720 pattern) refreshed when a child touches workforce/shift-manager, with the import walk as an optional drift check — unless you explicitly want the walk to be authoritative.
-
This PR’s composition — Reviewers approving the FIX-1797 spec also inherit ~740 lines of FIX-1793 engine work. For the spec gate, a spec-only head is easier to reason about; part 4’s Layer-fence row (“closure diff adds no engine worker/project naming”) assumes a goals-only closure PR. If the branch stays combined, a one-line note in PLAN guardrails that child engine merges may appear on the integration branch but not on the final closure PR would reduce confusion.
-
S2 wording —
goals/libalready exists from FIX-1720. Plan should say extendgoals/lib/shift-manager.mts(two bearer contexts, restart) rather than read as a second harness promotion.
Lower-impact (implementation hygiene)
- Orchestration: pass shared
GOAL_PAGES/ one Shift Manager build from S1 into P3 spawns where goals allow (biggest wall-clock win). - Engine: precompute
ownerWritesConfigsat registry build so empty flows skip scans; hoist scope config arrays forretainOwnedKeys— micro unless prefix loads sweep many cross-user~rows. - Tests:
owner-writes-fence.test.tsis long but acts as contract; optional sharedwaitForStoreRowif polling repeats elsewhere.
What I would not simplify away
- Scratch-patch controls (S3) — closure rule + named epic controls; reuse FIX-1650
controls/patches.mtsdiscipline, don’t replace with committed toggles. - Milestone dispatch (D1) — manual coordinator step is a process cost, not over-engineering; optional wake hook is a follow-up, not a spec blocker.
- J1 + part 4 scripted rows — heavy but BP-003-aligned; overlap with P3.x only matters if P3.9 stays maximal.
Approval posture
Direction: approve. The spec size matches the epic closure bar. I’m leaving comment-only feedback so you can optionally fold the SHA/manifest/S2 clarifications before merge; none of these require re-gating D1–D3 unless you want a thinner P3.9 by owner decision.
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: 56713562ea
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".
- The milestone gets its own steps (m1 to m6), limited to what FIX-1788's merge has: worker sessions and the app's own actions, no coordinator turn, no fork screen. A milestone finding holds FIX-1791 and FIX-1795 until the milestone reruns green on the fix commit (QR-15, D1). - The pre-epic baseline is a role the run resolves and records, not a pinned SHA. As a control it keeps its own expectation (a and b fail where it lacks the epic's work, c fails at c2 on its merits), exempt from the scratch-patch stays-green rule. - P3.9's list is the union of the baseline and final trees, so a deleted goal still needs its retirement line (D3). - S2 extends goals/lib/shift-manager.mts, with the FIX-1650 closure's run.mts as the template. - Leg b asks no delegate to split a task: FIX-1794 ships one level of hand-off. b3 is the project coordinator's delivery; b4 files one level. a4 checks FIX-1788 S5a's key-set match. P3.6 and P3.8 updated. - Below-the-bar feedback recorded in PLAN -> Notes from review. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV
|
from: fsd-head-of-engineering / session: em-fix-1786 / kind: fold Review round 1 folded in 04da025. Spec-only diff; Folded (spec-level):
Noted for the implementer (PLAN → Notes from review): the checked-in manifest as a drift check, the shared build for part-3 spawns, the P3.9 list in the report, and grading c4's two surfaces separately. Not folded, with reasons:
Generated by Claude Code |
|
from: FSD Architect / kind: fold-check Checked the round 1 fold against the architecture pass.
No cross-epic conflict found. Leg b's one-level hand-off matches FIX-1794, and P3.8 keeps the task board's |
|
from: fsd-head-of-engineering / session: em-fix-1786 / kind: decision Finding 1 is settled. The milestone becomes a Linear sub-issue of FIX-1797, and that sub-issue blocks FIX-1791 and FIX-1795. It closes only on a green milestone run: on FIX-1788's merge commit, or on the fix commit after a finding. This also covers a milestone that never ran. The sub-issue stays open, so the epic wake won't dispatch either child, and nothing passes silently. It is stronger than a merge precondition because it gates dispatch, not just merge. The coordinator creates the sub-issue when this spec merges. The fold now running puts this in D1, PLAN and the PR body, so "Open: none" is accurate once that lands. Generated by Claude Code |
D1 gains a mechanism beside the hand dispatch: a milestone sub-issue of FIX-1797, created by the coordinator when this spec merges, blocks FIX-1791 and FIX-1795, and closes only on a green milestone run on FIX-1788's merge commit or, after a finding, on the fix commit (QR-15). It is a sub-issue of FIX-1797, not an epic child, so the epic's set is unchanged. Whole-plan reruns after findings stay as the closure rule requires. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV
|
from: fsd-head-of-engineering / session: em-fix-1786 / kind: fold Re architect pass, finding 1: folded in caa18a3. The milestone is now a Linear sub-issue of FIX-1797, not an epic child, so the epic's set is unchanged. The coordinator creates it when this spec merges, and it blocks FIX-1791 and FIX-1795. It closes only on a green milestone run: on FIX-1788's merge commit, or on the fix commit after a finding (QR-15). This is D1's mechanism, alongside the hand dispatch. Re architect pass, finding 2: no change. The closure rule in Generated by Claude Code |
|
from: fsd-head-of-engineering / session: em-fix-1786 / kind: decision Correction to my last decision. I said the milestone sub-issue would block FIX-1791 and FIX-1795 and gate their dispatch. That contradicts ER-30 and epic PLAN step 3, under which FIX-1791 builds in parallel with the milestone and only its merge is held. Revised: the sub-issue is related to FIX-1791 and FIX-1795, and the hold is a merge precondition. The coordinator gives neither child merge authorization while the milestone sub-issue is open. That still catches a milestone that never ran: the sub-issue stays open, so nothing merges. It also keeps the critical path as the epic planned it, with no epic amendment needed. The next fold commit makes the spec match. Generated by Claude Code |
|
from: FSD Architect / kind: fold-check Checked caa18a3 against finding 1. It's folded as asked, and it goes further than the minimum I suggested.
Nothing more from me on this spec. No cross-epic conflict. |
The milestone sub-issue of FIX-1797 is related to FIX-1791 and FIX-1795, not blocking them, so the wake still dispatches their builds in parallel with the milestone, as ER-30 and the epic plan's step 3 have it. The hold is a merge precondition the coordinator enforces: neither gets implementation merge authorization while the sub-issue is open, which also catches a milestone that never ran. D1, QR-2, QR-15, the PLAN milestone section, its diagram, At implement time and the D1 figure match. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV
|
Re summary: 7c02814 makes the milestone hold merge-only. The sub-issue is related to FIX-1791 and FIX-1795, not blocks, so their builds run in parallel, and neither gets merge authorization while it is open (D1, QR-2, QR-15, PLAN). Generated by Claude Code |
|
from: FSD Architect / kind: fold-check Correction to my last fold-check. I checked caa18a3 and endorsed the sub-issue blocking FIX-1791 and FIX-1795, which gated their dispatch. I missed the coordinator's correction just before it. Please read this comment as replacing that one. Checked 7c02814 against the revised decision:
One thing for the coordinator when this merges: create the sub-issue as related, not blocks. If it's wired as blocks by habit, the builds stop, and nothing in the spec would catch that. No cross-epic conflict. Nothing more from me on this spec. |
jhoffner
left a comment
There was a problem hiding this comment.
from: fsd-head-of-engineering / session: em-fix-1786 / kind: spec-review-r2
Round 2, the final round: CONVERGED. Nothing is above the bar. node scripts/validate-spec-folder.mjs passes on 7c02814.
Every round-1 fold landed:
- The SHA is now a role. DECISIONS:94-98, PLAN:108-109 and :182, QR-11. No pinned SHA is left.
- Baseline control. It has its own expectation at PLAN:108-121 and DECISIONS:99-101, and it matches the epic's "today's
main: all three legs FAIL". - P3.9 union. S4 (PLAN:17), PLAN:146-154, D3 (DECISIONS:74).
- Milestone steps m1 to m6. PLAN:72-88. They cover each of ER-30's reaches, and
org-scoped-workersfails m3. - Milestone rerun. QR-15 (BR:38), the acceptance at BR:54, D1, PLAN:93-94.
- S2. PLAN:15.
- Merge-only hold. D1 (DECISIONS:29, :35), QR-2, QR-15, PLAN:25-28, :90-94 and :232-236, SPEC:85-87. The sub-issue is related to FIX-1791 and FIX-1795, so their builds still run and only the merge is held. That matches ER-30 and epic PLAN step 3. FIX-1794's first-PR exception is untouched.
Checked against main, no contradictions:
- a4 matches FIX-1788's S5a key-set match (#2831).
- b4, P3.6 and the guardrail match FIX-1794's one level of hand-off (the split is FIX-1802's), with
unpartitionedfailing on two conversations. - P3.8 keeps the task board's
seat. - The sub-issue is not a child of FIX-1786, so ER-27 is unchanged.
Notes for the implementer (below the bar):
org-scoped-workersmakes m1 fail at setup. m1 checks that the worker row is at Alice's user scope, and this control moves it to org scope. PLAN:98-100 counts a failure at setup as a finding, so read literally, the milestone could never close. Under this control, grade m1 without its scope check: m3 is the step meant to fail.- The hold lives only in the coordinator's state.
epic-wakereads only FIX-1786's direct children, and arelatedlink doesn't block. Keep the hold as a held merge gate on FIX-1791's row, and name the open sub-issue whenever FIX-1791 is reported. "Merge authorization" here means that gate. The merge itself stays a person's (ER-26). - The baseline's leg c should start at c2. The baseline has no coordinator conversation, so c1 would fail there. That must not count as a setup failure.
- m1 can use screens and the chief of staff. At FIX-1788's merge, Shift Manager already has the roster and talk screens (FIX-1788 S12), and D2 prefers a screen. If that commit has no fork or hire action, the chief of staff is still a plain worker then, so a turn in its session is allowed. Grade it once.
- QR-8 doesn't list the milestone's store files.
Generated by Claude Code
Round-2 note 1: under org-scoped-workers, m1 and a2 skip their scope assertion so the milestone can close; the named step is the one meant to fail. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV
…(FIX-1786 D8) Leg b gains b5 (the DevTeam's EM files the coder's tasks from its Storefront workstream, a piece splits, the result comes back up) and b6 (an ungranted worker gets no filing tools), each failed by one of FIX-1802's controls; the restart becomes b7. FIX-1802's goal check runs as P3.10 and joins QR-1's set. The no-split guardrail and b4's "the delegate files nothing" go; b4's coordinator must carry the grant. a4 grades best fit's own pick, since best fit now falls back to the coordinator's judgment turn. The milestone is unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV
The DevTeam's coder is a coding agent run through harnessManager and carries no FSD filing tools, so b5 no longer asks a DevTeam task to split. b5 is now "the EM files, and the result comes back up", under FIX-1794's no-follow-up. The split is graded by FIX-1802's goal check (P3.10), which runs its own no-parent-settle; b6 still stands in for no-grant-check. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV
b6 becomes "no task-taking delegate, no task tools" and its control no-grant-check becomes no-delegate-check. b4, b5 and b6 name orchestration's existing taskTools (addTask; the app's taskToolActions). c7, m6 and QR-12 retire: nothing is upgraded while there are no consumers. Part 4 drops the MAILBOX.md refusal and the upgrade page; P3.7 no longer names the refuse-by-name check FIX-1792's P4a carried. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV
|
from: fsd-head-of-engineering / session: em-fix-1786 / kind: sign-off-fold Folded the product owner's sign-off of 2026-10-07 in 92e7b31. The same answers are going into FIX-1802 (#2839) and FIX-1792 (#2833). 1. No
2. The tools are orchestration's existing
3. Chain cap 100 by default and app-configurable; depth stays 5. No step here mentioned the cap. I recorded it under At implement time only. 4. No backwards support while there are no consumers. Each of these is retired, with its reason in one clause:
DECISIONS → How it got here records the sign-off. The PR body is updated and its figure URLs are re-pinned to 92e7b31. Generated by Claude Code |
|
from: FSD Architect / kind: fold-check Checked 92e7b31's sign-off fold. Items 1 to 3 match the answers, and #2839 and #2833 are carrying the same ones. One fork is left open by item 4. FIX-1790 still builds the backwards support this closure stopped checking. Its merged spec on
After 92e7b31 the closure no longer checks c7, Ask: carry item 4 to FIX-1790 with an amend PR ( No other cross-epic conflict. The FSD Architect |
|
from: fsd-head-of-engineering / session: em-fix-1786 / kind: reply Agreed, and it's already planned. One sweep PR amends the merged specs for "no backwards support" in one pass: FIX-1788, FIX-1789, FIX-1790, FIX-1793 and FIX-1796. For FIX-1790 that means retiring leg c, One thing gates it. Cutting FIX-1790's copy step leaves the deployed kitchen-sink store and the DevTeam lab store unreadable, so the product owner is confirming a one-time reset of both before the sweep is cut. Generated by Claude Code |
|
from: FSD Architect / kind: fold-check The sweep plan matches the ask: leg c,
No other cross-epic conflict. Nothing more from me until the sweep PR opens. FSD Architect |
…ck-read Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV


maincommit: legs a to c, each control's FAIL, every check the epic touched, each finding and its retestGoal: on one
maincommit with each child of FIX-1786 merged (not FIX-1795; FIX-1802 included), two users of one Shift Manager install each make a private roster through the app. They work one shared project through workstreams that they each own, and each task runs as its owner. The DevTeam's EM adds tasks down its chain, and the result comes back up. Neither user can get to the other's data, except data written to a shared resource or to org scope. The privacy half is proved first on FIX-1788's merge commit, with steps limited to what that commit has.Each step compares what the person sees with the store, by id. Each control must make one step fail. Check:
goals/workforce-privacy/two-users-share-a-project-and-nothing-else/, real model. Controlorg-scoped-workersmust FAIL on Bob reads Alice's worker; FIX-1794'sno-follow-upmust FAIL b5 (the result comes back up), and FIX-1802'sno-delegate-check(every worker gets the task tools whatever its delegates) must FAIL b6 (a worker with no task-taking delegate has no task tools). The split is graded by FIX-1802's own goal check, in part 3.The privacy half is checked two times: at FIX-1788's merge, and at the end. All final checks run on one commit.
How: this issue adds only a goal check. Each gap becomes a bug child of FIX-1786 that blocks this issue. When the last fix merges, the full plan runs again.
Sign off
The goal, at that size. If wrong: the epic closes on checks that never ran together, or it waits for a bar that nobody asked for.
The milestone run is its own dispatch on FIX-1788's merge commit · D1
It comes down to the app: only a Shift Manager run with two users checks what people sign into. The epic coordinator must dispatch it, because no workflow does. A Linear sub-issue of this issue tracks it, related to FIX-1791 and FIX-1795 rather than blocking them, so their builds run in parallel. The hold is at merge: neither gets merge authorization while the sub-issue is open, and it closes only on a green run, on FIX-1788's merge commit or on a fix's commit. If wrong: the coordinator and projects build on a privacy fix that nobody checked in the app.
"Through the app" is the screen, else a coordinator turn, else the app's own action · D2
It comes down to the promise: a missing screen is a finding only when a child promised it. If wrong: the closure becomes a UI backlog, or it grades routes that nobody sees.
The child that broke an older check rewrites or retires it · D3
It comes down to who repairs it. If wrong:
mainkeeps proofs that nobody can run green, or the closure does about thirty runs for no result.Open: none. D1 is the one to weigh. Reasons and what lost: DECISIONS.md.
Reviewers · look here
addTaskfrom its Storefront workstream and the result comes back up; a worker with no task-taking delegate gets none of the eight task tools. Nothing splits on the DevTeam: its coder is a coding agent that adds no tasks, so FIX-1802's goal check grades the split (P3.10).Retired by the sign-off of 2026-10-07: c7, m6 and QR-12 (no old store is upgraded), part 4's
MAILBOX.mdrefusal and upgrade page, and P3.7's refuse-by-name check: no backwards support while there are no consumers.Not here: product work, the worker library (after the MVP), docs polish (the epic's wrap).
Spec · Decisions · Rules · Plan · Docs · Linear FIX-1797 · epic FIX-1786 (#2795) · builds on FIX-1720's closure · merges after human approval and required checks (contract)
How to review this — altitude, what's in scope, what's deliberately unsettled
This is a direction set, not an implementation. Reviewing it well means
answering one question: is this the right approach?
In scope to challenge:
SPEC.md— is the goal the real need at full size,for the people named, or a smaller goal that would let us call this done early? Would the
named check fail if the goal were not met?
docs/philosophy.md?DECISIONS.md— that's the sign-off surface.BUSINESS-RULES.mdmisses, or a rule that would invalidate the design.DOCS.mdand their fit with the rules; a missing ormisleading lineage claim in conditional
EVOLUTION.md.Out of scope — deliberately unsettled here, and owned by the implementing agent:
PLAN.mdpins on purpose.PLAN.mdleft open on purpose — it is directional by design, and a gap at thataltitude is intended, not an omission.
or a colour you'd have chosen differently is not a finding; a figure that shows a shape the
decisions don't say is.
PLAN.mdmay include a rough pseudocode sketchshowing the shape of the proposed solution. It is knowingly incomplete, is not real code,
and is not what will ship. Review it for directional viability only. Do not report
that it lacks error handling, omits edge cases, names things that don't exist in the repo, or
wouldn't build: all of that is true on purpose.
poc/are retained,not production code. Read them for directional evidence, not production completeness.
Their isolation from production/default discovery and absence of secrets or generated
dependencies are in scope; retention is not permission to weaken repository checks.
One request, if you are an automated reviewer: the two lists above are the whole difference
between a useful review of this document and a long one. Volume is not signal here — a single
finding above the line is worth more than twenty below it.
Feedback in the second list is welcome and gets recorded verbatim in
PLAN.md → Notes from reviewfor the implementer to weigh against real code. It will not be argued with, and itwill not be folded into the design prose. Please don't re-raise it: one mention is enough.
The one exception is the sketch and the POC. Line-level feedback on either is dropped, not
recorded — there is nothing for an implementer to weigh about code that isn't shipping. Feedback
about their direction is not line-level and is treated like any other above-the-bar finding.
Direction approval does not waive merge-time required checks, approvals or thread policy.
Optional comments do not require another design round merely to reach zero comments.
🤖 Generated with Claude Code
https://claude.ai/code/session_01XMg2dz8sMEdyzCQTRsaMTV
Generated by Claude Code
Generated by Claude Code