Repository navigation
spec(epic): W5 — Workforce release QA: Devtool, a DevForce proof, multi-seat collab (FIX-1457) - #1944
Conversation
The epic-spec set for W5: humans in seats, and a Workforce someone can actually run. Four documents and three figures under spec/_epics/living-workforce/. The objective: W3 made a Workforce describable and W4 is making work reach a seat. Neither produced something a person can run, and neither answered who the seat is when the seat is a person. W5 composes the two into runnable things, with humans occupying the same roster slot agents do — deliberately not a fourth substrate epic. The set is three spine points and two filed children. FIX-1458 (humans-in-seats exploration) is the one child sanctioned to run now. FIX-1455 is a held child epic with its own lifecycle. The third row, working assemblies, is unfiled and its cut is undecided until W4's first cut lands. Two things the gate should see: no filed child owns the epic's done condition (ER-19), and the assemblies row may collapse into FIX-1455 entirely. Both are recorded as open, not smoothed over. Also recorded: the end-state POC was skipped by decision, and W5 started as the fourth active epic against a cap of two (owner's call, 2026-09-19). Never merges. Open for the life of the epic. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016qaz1VpRK4wCQzxpjzZ4Hz
|
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. |
jhoffner
left a comment
There was a problem hiding this comment.
FSD Architect — coherence / approach stamp only.
Not owner approval. Not the merge gate. This is a never-merge epic PR; product sign-off stays with Jake on the Sign off surface (D2 / D4 / done-condition gap). Shared jhoffner login — do not treat this as Jake.
Two-direction check
No silent fork against in-flight W3 (#1718), W4 (#1905), or the FIX-1457 Linear spine.
- Sibling to W4, soft-after first-cut for ship (D4) matches locked sequencing. Explore-now / ship-fenced is the right split — not nesting under W4, not starting ship against a moving floor.
- Same seat slot, drain UX differs (D2) + assignee stays drain key / no new
TaskStatus(D3) match Architect invent-kills. No Human L1, no parallel HITL plane, no Lab-in-KS, Collab RC stays parked. - Composition, not substrate (D1) + feature-complete bar (W3 closed + W4 first-cut + org never optional + Labs without wrappers) matches the delivery-countdown finish line. Working assemblies compose W3 floor + W4 routing — they do not re-own boards, inventory, package cohesion, or dispatch honesty.
- FIX-1455 as held nested child epic (ER-15) and FIX-1458 as the only child sanctioned now match the reparent + explore cut. Unfiled assemblies + collapse trigger into KS is the right open, not a second demo stack invent.
Structural gaps already called correctly
Open 3/4 and ER-19 (no filed owner of the done condition) are the real gate questions — keep them open; do not paper them with a premature assemblies cut before W4 first-cut lands. D5 (fourth active epic vs cap of two) is recorded as a deliberate owner breach; fine for this cycle while live W5 work is a single exploration.
Out of scope here
FIX-1458's four walls (kind vs principal:, one↔many, channel members, durable human hire) stay exploration-owned. Issue-level approach belongs on child specs once filed.
— FSD Architect
| |---|---|---|---| | ||
| | [FIX-1458](https://linear.app/fixpoint-labs/issue/FIX-1458) · Explore: humans-in-seats | The **shape** of a human seat: the same roster slot, a Waiting-on-you drain, and the four open walls either answered or explicitly left open | Spine point 1, and the gate on the other two — nothing else in the set can be specced until a person in a seat has a shape | **Needs spec** · route `spec` (Design + Feature) · no PR yet · **the one child sanctioned to run now** ([D4](DECISIONS.md#d4)) | | ||
| | [FIX-1455](https://linear.app/fixpoint-labs/issue/FIX-1455) · Kitchen-sink rebuild | The reference consumer people copy: durable hire, `CHANNEL.md` seats, boards, inventory, inline vs resource-backed UI | Spine point 2 — the living app the objective is proved in rather than described in | **Held child epic** · own `Epic` label, own child [FIX-1429](https://linear.app/fixpoint-labs/issue/FIX-1429) (a `Bug`) · runs its **own** lifecycle ([ER-15](BUSINESS-RULES.md)) · ship fenced. No spec PR here **by design** | | ||
| | FIX-XXX · working assemblies | Reference surfaces composing the W3 floor and the W4 first cut into runnable product shape | Spine point 3, and the only place [ER-19](BUSINESS-RULES.md) — the done condition — can be proved | **Not filed** · the child cut is **undecided** until W4 first cut lands ([Open 3](DECISIONS.md#open)). One row standing for a spine point, **not** a promise of one issue | |
There was a problem hiding this comment.
Collapse leaves ownership undefined
This row says assemblies are the only place ER-19 can be proved, but the same spec allows the assemblies row to collapse entirely into FIX-1455, and DECISIONS.md says either row may own ER-19. On that collapse path, neither ER-19 nor ER-4 is reassigned. Coordinators therefore cannot tell whether to file redundant assemblies work or whether FIX-1455 provides a valid path to wrapping the epic. Make the ownership and dependency graph conditional, or define the collapsed ownership explicitly.
| 4. **[D5](DECISIONS.md#d5) · W5 starts as the fourth active epic, against a cap of two.** Your call | ||
| on 2026-09-19, recorded so a later reader knows the breach was deliberate. If wrong: four epics | ||
| all move slowly instead of two moving fast. |
There was a problem hiding this comment.
Sign-off presents wrong decisions
The repository's decision-framing guide caps the sign-off surface at three live asks and requires an already-made call to be reported as news rather than offered for approval. This section instead includes four items, uses one slot to re-ratify the already-decided D5 breach, and omits D3 and the unresolved done-condition ownership fork. Move D5 out of Sign off and use this surface for the decisions approval must actually settle. This repository requirement must be satisfied before merging.
Context Used: CLAUDE.md (source)
Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!
| <text class="t-use" x="664" y="256">a gap, not an owner</text> | ||
| <text class="t-use" x="16" y="284">Only one column is live this cycle. FIX-1455 runs its own epic lifecycle; the assemblies column is not filed at all.</text> | ||
| <text class="t-use" x="16" y="302">ER-19 is the epic's done condition and nothing owns it, because the row that would has not been cut — Open 3 and 4.</text> | ||
| <text class="t-use" x="16" y="320">The other nine rules are prohibitions, which bind every row equally and so have no column to sit in.</text> |
There was a problem hiding this comment.
Ownership summary miscounts rules
The figure says the remaining nine rules are prohibitions, but BUSINESS-RULES.md contains ten prohibitions, four process rules, and two additional proof rules outside the five-row matrix. The visible summary is therefore incomplete and differs from the model conveyed by the aria-label, which does not explain the omitted rules. Update both descriptions to state accurately which 16 rules are intentionally outside the matrix.
There was a problem hiding this comment.
Simplify review (epic-spec · FIX-1457)
Parallel read-only passes over the four documents + figures. No implementation scope — feedback is on set shape, ceremony, and whether a cleaner decomposition exists.
Verdict
Approve the direction. D1 (composition, not a fourth substrate epic), D2 (human in the same seat slot), D3 (assignee ≠ seat; no new TaskStatus), and D4 (explore now / ship after W4 first cut) are the simple choices relative to the rejected forks in the decision tree. Skipping an epic-altitude end-state POC and pointing that work at FIX-1458 is also the right cost trade.
What adds coordination surface is mostly organizational, not wrong product calls — and several items are already named in the spec (collapse trigger, Open 3/4, ER-15).
Where the approach may be heavier than necessary
1. Three spine points vs two (largest simplify fork)
The set table, mermaid graph, ownership figure, ER-4, and Opens 3–4 all maintain a third row (FIX-XXX · working assemblies) while Open 2 already leans toward FIX-1455 being spine point 3. That is honest per the template, but it multiplies “no owner for ER-19” prose across SPEC, DECISIONS, BUSINESS-RULES, PLAN, and sign-off.
Cleaner default hypothesis (optional at gate, not a blocker): Narrate W5 as two children — FIX-1458 (shape) + FIX-1455 (reference consumer and provisional proof owner for ER-19). File a assemblies issue only if W4’s first cut proves work that cannot live in FIX-1455’s lifecycle. You already wrote the collapse trigger; consider stating that hypothesis above the table so readers do not treat three rows as the committed plan.
2. Nested FIX-1455 + third row
ER-15 is well argued if FIX-1455 stays an Epic. The smell is listing FIX-1455 as a set row while also saying “2 filed children” and running a second lifecycle — coordinators must track two epic PRs, an empty spec-PR cell, and wrap ordering. Alternatives if coordination cost bites later (not asking to rewrite now): list FIX-1455 under “related epic (not in this set)” with one ER-15 pointer, or fold KS rebuild as a large W5 child if product scope ever allows one gate.
3. Ceremony / duplicated state (doc hygiene)
- Ship fence (D4): same invariant in D4, ER-10, SPEC box copy, PLAN path, and both lane figures. Define once in D4; elsewhere link.
- Prohibitions: ER-5–14 overlap D1–D3;
end-state.svgaria-label restates much of the NOT BUILT strip. Consider keeping the long list in BUSINESS-RULES only and thinning figure aria-labels to pointers. - PLAN “Where it is”: per-ticket W4 enumeration will stale; W4’s epic set table is the reuse target (same pattern as ER-16 here).
- Sign-off: D3 is as consequential as D2 for children (ER-7/8) but is not in the numbered sign-off list — either add it or one line that D3 is consumed from W4 with no re-gate.
- D5: process/capacity metadata; fine to record, but it is not a product fork — could live only in “How it got here” unless you want every gate to re-approve capacity.
4. Process / coordination cost (not runtime)
D5’s four active epics against a cap of two is a real parallelism tax for the whole W5 life, even if this cycle is only FIX-1458. ER-20-style “cross-spec before any child ships” + FIX-1455 outside this spec PR risks two review planes unless orchestration batches once — worth one sentence in PLAN or ER-20 on how parent/child cross-spec compose.
What I would not simplify away
- FIX-1458 as the only lane crossing the ship fence this cycle.
- “Is three really two?” + explicit collapse trigger.
- ER-19 gap flagged at sign-off (live fork in the PR body) rather than papering over with a fake owner.
- Coordination seams table (1458↔1455, FIX-1430, FIX-1442) — high signal, low word count.
Reuse check
Form matches epic-spec-template and W4 (work-routing-package-cohesion): dated set table, fence vocabulary, placeholder row, related-not-children for W4-side work. W5 is leaner in total word count than W4’s DECISIONS ledger — good. Proof pattern reuse from W4 (manager-queue lab / second goal check) fits ER-19 better than a third “assemblies” epic child if human drain is an incremental goal on the same lab or KS path — aligns with collapse to two children.
Skipped (needs product call, not simplify nit)
- Forcing FIX-1455 out of Epic into a single W5 issue.
- Pre-emptively deleting the assemblies row before W4 first cut lands (your Open 3 is the right fence).
No changeset; spec-only PR.
Sent by Cursor Automation: Review For Best Approach
| |---|---|---|---| | ||
| | [FIX-1458](https://linear.app/fixpoint-labs/issue/FIX-1458) · Explore: humans-in-seats | The **shape** of a human seat: the same roster slot, a Waiting-on-you drain, and the four open walls either answered or explicitly left open | Spine point 1, and the gate on the other two — nothing else in the set can be specced until a person in a seat has a shape | **Needs spec** · route `spec` (Design + Feature) · no PR yet · **the one child sanctioned to run now** ([D4](DECISIONS.md#d4)) | | ||
| | [FIX-1455](https://linear.app/fixpoint-labs/issue/FIX-1455) · Kitchen-sink rebuild | The reference consumer people copy: durable hire, `CHANNEL.md` seats, boards, inventory, inline vs resource-backed UI | Spine point 2 — the living app the objective is proved in rather than described in | **Held child epic** · own `Epic` label, own child [FIX-1429](https://linear.app/fixpoint-labs/issue/FIX-1429) (a `Bug`) · runs its **own** lifecycle ([ER-15](BUSINESS-RULES.md)) · ship fenced. No spec PR here **by design** | | ||
| | FIX-XXX · working assemblies | Reference surfaces composing the W3 floor and the W4 first cut into runnable product shape | Spine point 3, and the only place [ER-19](BUSINESS-RULES.md) — the done condition — can be proved | **Not filed** · the child cut is **undecided** until W4 first cut lands ([Open 3](DECISIONS.md#open)). One row standing for a spine point, **not** a promise of one issue | |
There was a problem hiding this comment.
Simplify · default to two children? This row plus FIX-1455 makes three spine surfaces (table, mermaid, ownership SVG) while Open 2 in DECISIONS already leans toward KS being spine 3. Consider leading the set table with the collapse hypothesis (two children; assemblies filed only if 1455’s set cannot host ER-19) and keeping FIX-XXX as a one-line “file when W4 lands if needed” note instead of a full third row — if you want less gate churn on Open 3/4. Not blocking; the written collapse trigger is the right escape hatch.
| and its first cut has **not** shipped, so two of the three rows are fenced and the third is the | ||
| whole of this cycle's W5 work. | ||
|
|
||
| **Is three really two?** The assemblies row is the one that can collapse: the locked input already |
There was a problem hiding this comment.
This paragraph is doing the right work — keep it. If you adopt a two-child default narrative, move it above the table so readers do not treat three rows as committed.
|
|
||
| ## What the end-state POC showed | ||
|
|
||
| **None was built, and that is a decision rather than an omission.** FIX-1458 *is* the humans-in-seats |
There was a problem hiding this comment.
Simplify win. Skipping epic-altitude end-state POC and tying revisit to assemblies cut avoids duplicate exploration cost; FIX-1458 as the shape vehicle is the cleaner approach.
|
|
||
| Four, plus the exploration's own walls. | ||
|
|
||
| 1. **Nothing in the set names its proof surface.** ER-19 describes what done looks like; no artifact |
There was a problem hiding this comment.
Opens 1, 3, and 4 restate the same structural hole (ER-19 / proof surface / owner). Consider merging 1 into 3–4 or assigning provisional owner: FIX-1455 in ER-19 with “unless assemblies cut says otherwise” — would shorten sign-off without hiding the gap.
|
|
||
| | # | The epic is done when | Proved by | | ||
| |---|---|---| | ||
| | ER-19 | A **person occupies a seat** in a running reference assembly, is assigned a board row by a coordinator, and drains it through Waiting-on-you — observed on the real path, not described | **No owner yet.** The row that would own this is not cut ([Open 3 and 4](DECISIONS.md#open)). This is the epic's one structural gap, and the sign-off should treat it as one | |
There was a problem hiding this comment.
Flagging no owner is correct for the gate. Optional simplify: pair with provisional 1455 owner in the same cell so cross-spec reviewers have a default place to look for proof work (W4’s ER-20 → FIX-1430 pattern).
| Status lives in one place: [the set table in the spec](SPEC.md#the-set--as-of-2026-09-19). The lanes | ||
| above carry the same state as a picture of time and are redrawn when it moves. | ||
|
|
||
| **The inputs from other epics, and their verified state as of 2026-09-19:** the **W3 floor** |
There was a problem hiding this comment.
Reuse · avoid stale laundry list. This W4 ticket roll will drift. Prefer linking W4’s epic set table / fence state (same mirror pattern as ER-16) and one line: “W4 first cut (FIX-1407): not shipped.”
| If the cut names nothing outside FIX-1455's set, the row **closes** instead and W5 is two children. | ||
| 4. **ER-19 is observed** → the epic wraps. | ||
|
|
||
| ## Coordination seams to watch |
There was a problem hiding this comment.
High-signal, low ceremony — the seams table is worth the words. If ER-19 proof reuses manager-queue lab goals (W4 style), note that here to steer away from a third demo stack when the assemblies cut is made.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1d9218e877
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".
|
|
||
| | Issue | What it delivers | Why the set needs it | Status | | ||
| |---|---|---|---| | ||
| | [FIX-1458](https://linear.app/fixpoint-labs/issue/FIX-1458) · Explore: humans-in-seats | The **shape** of a human seat: the same roster slot, a Waiting-on-you drain, and the four open walls either answered or explicitly left open | Spine point 1, and the gate on the other two — nothing else in the set can be specced until a person in a seat has a shape | **Needs spec** · route `spec` (Design + Feature) · no PR yet · **the one child sanctioned to run now** ([D4](DECISIONS.md#d4)) | |
There was a problem hiding this comment.
Resolve walls that block downstream specs
This lets FIX-1458 finish with any of its four walls merely “explicitly left open,” while PLAN.md:41-47 treats the resulting seat shape as the prerequisite that releases FIX-1455 and the assemblies cut. Because those walls include the human-seat identity model and durable-hire persistence (DECISIONS.md:122-126), leaving either open prevents the downstream specs from choosing a coherent contract even though their prerequisite is marked complete. Require resolution or owner sign-off for every downstream-blocking wall; only questions proven irrelevant to those children can remain open.
Useful? React with 👍 / 👎.
|
|
||
| | # | The epic is done when | Proved by | | ||
| |---|---|---| | ||
| | ER-19 | A **person occupies a seat** in a running reference assembly, is assigned a board row by a coordinator, and drains it through Waiting-on-you — observed on the real path, not described | **No owner yet.** The row that would own this is not cut ([Open 3 and 4](DECISIONS.md#open)). This is the epic's one structural gap, and the sign-off should treat it as one | |
There was a problem hiding this comment.
Add org identity to the completion gate
ER-19 lets the epic finish after proving only the human-seat assignment and drain path, but the objective explicitly promises a living reference that runs with org identity (SPEC.md:15) and defines the feature-complete bar as “org never optional” (DECISIONS.md:32). Because PLAN.md:57,64 treats FIX-1442 as soft-related rather than a wrap dependency, W5 can satisfy ER-19–ER-21 and close while that part of its objective remains unproved. Include org-bound execution in ER-19 or make the supplying work a real wrap dependency.
Useful? React with 👍 / 👎.
| | # | The epic is done when | Proved by | | ||
| |---|---|---| | ||
| | ER-19 | A **person occupies a seat** in a running reference assembly, is assigned a board row by a coordinator, and drains it through Waiting-on-you — observed on the real path, not described | **No owner yet.** The row that would own this is not cut ([Open 3 and 4](DECISIONS.md#open)). This is the epic's one structural gap, and the sign-off should treat it as one | | ||
| | ER-20 | Nothing the set shipped added an L1 type, a status enum value, or a second work plane | A cross-spec review over the set's specs before any of them ships | |
There was a problem hiding this comment.
Verify shipped code rather than only its specs
The stated condition concerns what the set actually shipped, but its only proof is a cross-spec review, which runs on approved specifications before implementation. If an implementation later introduces an L1 type, status value, or second work plane contrary to its spec, this check still passes and the epic can wrap with ER-20 false. Add a wrap-time check over the shipped child diffs or another implementation-level verification path.
Useful? React with 👍 / 👎.
| <text class="t-fen" x="528" y="84">W4 first cut → fence lifts</text> | ||
|
|
||
| <g id="critical-path"> | ||
| <path class="crit" d="M190 146 L190 190"/> |
There was a problem hiding this comment.
Route the proof path through the reference app
The figure draws the proof’s critical path directly from FIX-1458 to the assemblies lane, omitting FIX-1455, even though SPEC.md:61-63 routes FIX-1458 through FIX-1455 before assemblies and PLAN.md:25-26 says assemblies consume the reference app that supplies the ER-19 proof surface. A coordinator using this path index can therefore treat the proof as ready while the required runnable reference is still held; route the path through FIX-1455 or depict both prerequisites converging on assemblies.
AGENTS.md reference: AGENTS.md:L52-L52
Useful? React with 👍 / 👎.
jhoffner
left a comment
There was a problem hiding this comment.
second-look: PR #1944 (W5 epic-spec)
Verdict: Appropriately scoped — no overbuilt/YAGNI finding here clears this skill's calibration bar.
Scope: spec-only, +612/−0 across 7 files (4 markdown docs + 3 SVG figures, all under spec/_epics/living-workforce/). No code is touched by this PR.
| # | Finding | Lens | Recommendation | Est. Δlines | Sev | Conf |
|---|---|---|---|---|---|---|
| — | none survived the gate | — | — | — | — | — |
Two candidates got real scrutiny and were dropped rather than force-fit into the table (detail below): whether the 3-child set should collapse to 2 now, and whether D5 needs full decision-card treatment. Both are noted inline on the diff for whoever picks this up, but neither clears "net removal worth the row" on its own.
Architecture-improvement check (per improve-codebase-architecture's lens)
None to flag. Every changed file is a spec document or a hand-authored SVG figure — there is no source code in this diff to deepen, consolidate, or make more testable. This check only applies when a PR is also touching code it's shipping; this one is a pure coordination artifact that never merges.
Considered & deemed justified
- The unfiled "assemblies" row (spine point 3) kept as a placeholder rather than cut now — the collapse trigger depends on facts (what FIX-1455's actual scope becomes) that don't exist until the W4 first cut lands. Forcing the call now trades a real "we don't know yet" for a guess. See inline comment on
SPEC.md. - 21
BUSINESS-RULES.mdentries for a set with one active child — matches the epic-spec-template's four required subsections exactly (owned rules, prohibitions, how-the-set-is-run, done-condition), and the prohibitions (ER-5–ER-14) are guardrails for FIX-1455 and the assemblies work before they start, which is the entire point of deciding cross-cutting constraints at epic altitude instead of letting each child re-litigate them. SPEC.md(~45% over its ~900-word template budget) andDECISIONS.md(~18% over ~1,300) — self-flagged in the PR's own "Engineering calls" section. I looked for compressible content in both and didn't find any that doesn't cost a child information it would otherwise have to re-derive; the overage is concentrated in the two required tables, as claimed.- Three figures (box, ownership matrix, path) — exactly what
docs/contributing/spec-figures.mdmandates at epic altitude; not a choice available to trim. - D5's full decision-card treatment (portfolio cap breach, not a technical cross-cutting call) — real but too small to clear the bar on its own. Flagged inline as a minor, non-blocking note rather than a table row.
Two inline comments posted (on SPEC.md and DECISIONS.md) answering the PR's own "Reviewers · look here" #1 and noting the D5 observation, for whoever acts on this next.
Generated by Claude Code
| **Is three really two?** The assemblies row is the one that can collapse: the locked input already | ||
| leans against a third demo stack. **Collapse trigger:** an assemblies cut whose every item is | ||
| already a FIX-1455 issue — then the row closes and W5 is two children. The reverse risk is live | ||
| too, and it is [Open 4](DECISIONS.md#open): today **no filed child owns the epic's done | ||
| condition**, because the child that would has not been cut. |
There was a problem hiding this comment.
second-look, re: "Reviewers · look here" #1 — is the set two or three?
Ran this through the restraint/YAGNI lens explicitly, since it's the question the PR asks reviewers to answer. My read: don't pre-empt the collapse now. The evidence needed to answer it (what FIX-1455's own scope turns out to be once the W4 first cut lands) doesn't exist yet, so cutting the row today would be a guess dressed as a decision — and if the guess is wrong, ER-19 (the done condition) ends up structurally homeless in the wrong direction. Keeping it as an undecided placeholder, with the collapse trigger written down, is the correct amount of ceremony for "we don't know yet, and here's exactly what would tell us."
I did estimate what collapsing now would save across the four docs (the third set-table row, part of the mermaid graph, ER-4, and Open items 1/3/4) — it's on the order of 15-20 lines, which doesn't clear this skill's bar for a finding on its own, and forcing the decision early would be arguing timing, not cutting real bloat. So: no action requested here, just confirming the deferral reads as the right call rather than something this pass is flagging.
Generated by Claude Code
| <a name="d5"></a> | ||
| ## D5 · W5 starts as the fourth active epic, against a cap of two | ||
|
|
||
| | | | | ||
| |---|---| | ||
| | **Instead of** | Forcing W3 ([#1718](https://github.com/fixpoint-labs/flow-state-dev/pull/1718)), W4 ([#1905](https://github.com/fixpoint-labs/flow-state-dev/pull/1905)) or LAB-162 ([#1612](https://github.com/fixpoint-labs/flow-state-dev/pull/1612)) to wrap to free a slot | | ||
| | **Because** | The board was already at three open epic PRs against a cap of two when W5 was opened. The owner chose the breach on 2026-09-19 rather than wrap an epic early — W5's live work this cycle is a **single exploration**, which is close to free, and forcing a wrap would have cost real closure quality on an epic that isn't finished | | ||
| | **Locks in** | The cap is knowingly breached, not forgotten. If W5 ship work starts before two of the four wrap, that is a **second** decision and this card does not cover it | | ||
|
|
There was a problem hiding this comment.
second-look, minor/non-blocking: D5 gets the same full card treatment (tree node + rejected alternative + Instead-of/Because/Locks-in table) as D1–D4, but unlike those it drives zero BUSINESS-RULES.md entries and isn't on the sign-off ask (the PR body's own "Not asked" note says it's already decided). It's a portfolio-scheduling record, not a cross-cutting call any child must obey.
Checked this against the calibration bar before flagging: collapsing it to a one-line note under "Decided in review" (or leaving it only in the PR body's "Not asked" section) would save roughly 10-12 lines here plus the SPEC.md sign-off item — a "handful," not enough to clear this skill's net-removal threshold as a real finding. Noting it because it's the one place in the doc set where the ceremony slightly outruns what the content needs, not because it's worth a re-edit on its own.
Generated by Claude Code
Four reviewers on head 1d9218e (greptile, cursor, codex, second-look). - ER-19 gains a provisional owner: FIX-1455, unless the assemblies cut says otherwise; ER-4 travels with it. The collapse path had no defined owner. The three-row set is NOT collapsed to two — that was proposed and rejected the same round. - ER-18 gains a downstream-blocking clause: FIX-1458 does not report complete with the identity model or durable-hire persistence still open, absent owner sign-off. - ER-19 gains org-bound execution; FIX-1442 becomes a wrap-gating input rather than merely soft-related. - ER-20 gains a wrap-time sweep of the shipped child diffs. - SPEC.md's sign-off drops D5 (news, decided 2026-09-19, not an ask) and notes D3 as consumed from W4 with no re-gate. Three live asks. - ownership.svg: ER-4/ER-19 cells become conditional; the legend's rule count corrected from "nine prohibitions" to the sixteen outside the matrix. path.svg: critical path now stops at FIX-1455; now-line and set-table anchor roll to 2026-09-20. - Opens renumbered 4 -> 3 (old Open 1 is answered provisionally). Both figures rendered and checked in light and dark. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016qaz1VpRK4wCQzxpjzZ4Hz
…es (FIX-1457) ER-19 gained org-bound execution in the round-1 fold, but SPEC.md's sign-off fork still described the finish line as "a running app". A reader approving the fork was approving the weaker condition. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016qaz1VpRK4wCQzxpjzZ4Hz
|
spec approved |
…landed (FIX-1457) Status only; no objective, decision or rule moved. - Set table: FIX-1458 Needs spec -> Spec in review (#1955); counts line now 0 done / 1 spec in review / 1 held child epic / 1 not filed. - W4's first cut has landed (every first-cut child Done; FIX-1460 and FIX-1461 are backlog strays outside it), so the ship fence has lifted. Recorded on every surface that asserted the old state: the set table's FIX-1455 and assemblies rows, PLAN.md's dated input paragraph and its unblock list, and both figures. Nothing downstream of the fence has been started, and the assemblies cut is still Open 2/3. - path.svg: FIX-1458 gets a live in-flight bar across the now line, the W4 lane a solid landed bar, FIX-1455 an unfenced not-started line; the forward fence-lift marker is gone and the band reads LIFTED. - end-state.svg: the fence band and the consumed-W4 tile carry the new state. Both figures rendered and checked in light and dark. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016qaz1VpRK4wCQzxpjzZ4Hz
…stays (FIX-1457) Owner-directed decision flip, not review feedback and not a review round. D2's prior lean (a person occupies the same seat slot an agent does) is superseded as of 2026-09-20; the card records it as a flip. Model B: a non-human seat owns the task and parks it with a reason; the human acts through flow actions bound to a principal; the flow decides what the input means and continues. The org chart survives and is the point, with a durable bind. `flow: human` as a drain kind is demoted to Lab chrome. Surfaces re-derived with it (tenet 5): - ER-1 and ER-21 rewritten; ER-2 and ER-3 rephrased off the seat-drain spine - ER-6 also forbids a person as a board drainer - ER-18's two downstream-blocking walls renamed to the principal bind plus action authorization, and what a redeploy must round-trip to keep it — a rename, not a new clause; the clause's force is unchanged - ER-19 rewritten: the old done condition asked a person to DRAIN a row, which Model B forbids — the epic could have wrapped on the removed mechanism - FIX-1455 no longer inherits a human-drain-seat store requirement - D1's and D3's falsifier/lock-in wording taken off the seat model (D3 itself is unchanged and not re-gated) - box figure redrawn; ownership ER-1 row and path's manager-queue lane redrawn; all three rendered in both themes and checked - the exploration's four walls re-cut; the unparkAndDrain caller question recorded as NARROWED, not closed The objective gate was taken on 2026-09-20 and D2 was flipped after it, so the gate is re-taken on this version. No approval label applied or removed. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016qaz1VpRK4wCQzxpjzZ4Hz
|
FSD Architect — owner call (Jake 2026-09-20). FIX-1458 / #1955 out of W5. Explore canceled; PR closed unmerged. Revisit when multi-user pain is real. Invent-kill that stays on this epic: humans are not board drain seats; need-human = agent owns row → park → flow action with request principal; no Human L1 / parallel HITL plane. W5 ship cut = kitchen-sink + living assemblies (one-user-org / agent isolation). Multi-human org chart = grow-into later. EM: do not pick up humans-in-seats from this epic. Linear FIX-1457 body patched. — FSD Architect |
…-1457) Owner re-cut on 2026-09-20, in two steps. Not review feedback and not a review round: reviewRounds stays at 1. 1. FIX-1458 canceled, #1955 closed unmerged. Its invent-kills survive as a locked constraint (D2, no longer a fork); its product half is deferred with a revisit condition, not left Open. Boundary recorded as D7: one user, one org, agents isolated from each other. 2. Objective restated: polish and prove Workforce so it can ship, on three surfaces — kitchen-sink in action, devtool showing a run unfold across channels and workers, and a basic DevForce configuration. Recorded as D6. ER-19 rewritten a third time, now as three named artifacts each naming its producer. Writing it that way is what exposed that two of the three have no child: the devtool view and a basic DevForce are named gaps in the set table with no ticket and no owner. Stated rather than discovered at wrap. ER-18's two downstream-blocking walls were FIX-1458's and are retargeted to FIX-1467 (the noun and migration; the grant model). ER-1 and ER-2 moved to FIX-1455. No ownerless blocking clause left standing. FIX-1467 added to the set table, which had never carried it. The vague "working assemblies" row is resolved into the devtool and DevForce gaps. ER-22 (no multi-human machinery) and ER-23 (the set is open) added. ER-11 narrowed: DevForce as evidence is in, Lab product inside kitchen-sink is still out, CyberForce stays out. Salvaged from the canceled explore so it does not die with #1955: a session belongs to one user, so a board answered into cannot be session-scoped; the ledger is org-scoped. All three figures redrawn and rendered in both themes. The box figure draws the two unowned surfaces as empty slots; the ownership matrix draws them as NO CHILD cells; the path gives them lanes with no bar. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016qaz1VpRK4wCQzxpjzZ4Hz
|
FSD Architect — W5 recalibration (Jake 2026-09-20, FSD Prod/Eng). Cut: Workforce release QA / polish. Exit proofs (named):
Out of W5: kitchen-sink → FIX-1455; humans-in-seats stays canceled. Soft-after W4 first-cut for ship still stands. Linear FIX-1457 body rewritten for EM. Not owner merge approval on this never-merge epic PR — direction for the EM. — FSD Architect |
…fs (FIX-1457) Owner recalibration 2026-09-20 18:05Z. The epic's identity changed, not just its set. - ER-19 retired; replaced by ER-Devtool / ER-DevForce / ER-Collab, with the six-row Devtool checklist drafted against what the surfaces expose today. Two of its six rows fail as written. - D8 (new): proof-via-DevForce is IN — the invent-kill against it is dead. Thinnest path that produces a real artifact, then stop. - D9 (new): kitchen-sink left the set. ER-15 rewritten, ER-24/ER-25 added, and ER-1/ER-2/ER-3 re-owned so no rule points at a row that is gone. - ER-18 recorded as MET by FIX-1467's merge, not left standing open. - D4 corrected: the ship fence STANDS (FIX-1407 is In Review); polish may start. - FIX-1468/FIX-1469 entered the set table, both marked not a proving leg; FIX-1469's kitchen-sink half flagged stale. - Sign-off cut to one ask: whether ER-Devtool collides with FIX-1320. - All three figures redrawn and checked in both themes. Not review feedback and not a review round: reviewRounds unchanged. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016qaz1VpRK4wCQzxpjzZ4Hz
…the owner's answers (FIX-1455) Move the set from spec/_epics/kitchen-sink-rebuild/ to specs/epics/FIX-1455/ under the retained-spec policy. Relative depth is unchanged, so ../../../docs paths still resolve; scripts/validate-spec-folder.mjs passes. Add the two documents the policy now requires of this set: - DOCS.md — the shared reader narrative (a Workforce app's roster, channels and importable chrome), drafted prose for a new durable-hire page and a components page, and per-issue publication ownership. FIX-1429 and FIX-1476 carry justified no-doc-impact statements. - EVOLUTION.md — four predecessors: the epic body's CHANNELS.md lock (superseded), its board v1 lock (retained whole), its inline-as-lifetime vocabulary (amended), and W5's membership claim (superseded by the owner's FIX-1457 recalibration, PR #1944 D9). Fold the three answers the owner approved on PR #1978: - FIX-1475's durability proof is runtime hire only. ER-2 and ER-22 amended; ER-16 unchanged and FIX-1415 stays parked and out of the set. - The five stale tickets split: FIX-1372 stays open and is verified at wrap; FIX-420, FIX-472, FIX-429 and FIX-540 close as superseded. Documents only — Linear triage is PM/LM's. - Figure altitude trimmed: shell-before-after.svg moves beside the D7 card, SPEC.md's shell section is ~75 words, and PLAN.md records that FIX-1477 owns the regions narrative and inherits the three retained shell figures. No figure deleted. Reconciled surfaces that restated a superseded answer: SPEC.md's set table row for FIX-1475, the sign-off's open block, PLAN.md's FIX-1475 and FIX-1477 rows and its wrap, and figures/path.svg's held-pending-triage note (redrawn and rendered in both themes). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01ANuyMVi3YS653fcghb9gtB
Brings the retained-spec contract (#1975) and the updated spec-folder guard onto the epic branch ahead of migrating W5's spec to specs/epics/FIX-1457/. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016qaz1VpRK4wCQzxpjzZ4Hz
Moves the set from spec/_epics/living-workforce/ to specs/epics/FIX-1457/, keyed by epic issue id, with git mv so history follows. Adds the two documents the new contract requires: - DOCS.md — the one shared promise ER-21 makes, FIX-1481's three Devtool readings, and a plain statement of what cannot be drafted yet because two of three proofs have no producer. - EVOLUTION.md — five cross-cutting predecessors with precise anchors: ER-19's three-surfaces done condition, the DevForce invent-kill, W5's parenthood of kitchen-sink, D4's fence read as lifted, and ER-18 recorded as met. Cited by commit/PR/Linear provenance, since none has a retained artifact under specs/. The epic PR now merges after the objective gate rather than staying open for the life of the epic. PLAN.md's wrap said "close the epic PR unmerged"; it now derives completion from Linear and the implementation PRs and routes material post-merge changes to a follow-up PR from main. Set table: adds FIX-1481 (Devtool child, ER-Devtool producer for rows 4-6; rows 1-3 ride FIX-1320) and FIX-1474 (soft-related polish, explicitly not an ER-Collab gate). Both were already parented under FIX-1457 in Linear. Every surface that restated "zero proof-producers" is re-derived with them: the counts line, the dependency graph, Sign-off 1's outcome, ER-2's owner, ER-Devtool's and ER-Collab's proved-by cells, D5's arithmetic, Open 1, and all three figures (rendered and checked in both themes). No decision, rule, invent-kill or scope changed in the move. The Workforce release QA re-cut stands as it was. Verified: node scripts/validate-spec-folder.mjs exits 0 (it failed with "Found 1: spec/_epics" before the move); no spec/_epics or living-workforce citation remains outside one historical mention; all internal links and cross-document anchors resolve; spec-figures.md's six verify checks pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016qaz1VpRK4wCQzxpjzZ4Hz
'Round 1 exposed first two, and now all three, unowned legs' read as a present-tense claim and contradicted Open 1 after FIX-1481 was filed. It now records what round 1 exposed and states that two legs are still unowned. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016qaz1VpRK4wCQzxpjzZ4Hz
…owned (FIX-1457) Post-merge amendment to the retained epic spec at specs/epics/FIX-1457/, on a new PR from main. #1944 stays the merged review record and is not touched. Three changes, and one of them is a correction: - ER-DevForce and ER-Collab stop reading "No producer". FIX-1496 (In Spec Review, #2023 @ 25e7507) and FIX-1497 (Backlog, no spec) take them. FIX-1496's artifact cut is recorded in the epic's own words: one automated leg pushing to a bare clone at a declared path, resolving after the goal's process exits with no gh, no token and no network; the credentialed PR leg documented as the human release run. Channels are recorded as inside the ratified criterion, not optional. - ER-Devtool's row split is corrected from three-and-three to two-and-one-and- three. FIX-1481's spec deferred checklist row 5 on the claim that FIX-1477 PR-C already owned the reader. That claim was false: FIX-1477's Roster reads workforce/roster/*, FIX-1475's collection, while row 5 names inventory/*. Row 5 went to FIX-1502, blocked by FIX-1486. Recorded in EVOLUTION.md as a correction, with how it was caught. - ER-26 is new: no exit proof's run is graded before FIX-1496 stands a live hired Workforce up. It fences the runs, explicitly not FIX-1481's or FIX-1502's code PRs. Row 6 is the sharp case — no tree in the repository has a sealed document today. Open 1 is replaced rather than answered: the staffing gap closed, and whether W5 exits on five checklist rows of six is now in front of the owner, six-part form in DECISIONS.md. All three figures redrawn and rendered in both themes; no text overflows, measured in-browser rather than estimated. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016qaz1VpRK4wCQzxpjzZ4Hz
The stored failure is backticks wrapped around a raw.githubusercontent.com URL, including the attribute's closing quote. That was #1905 at create. Create and update both kept a clean img line on #1944 and on a 2026-09-23 remeasure, so the create-versus-update rule was sending refreshes to drop figures that would have rendered. Read the stored body back. Keep a src with no backtick. Hand a person the clean line only when the src comes back defanged. Co-authored-by: Jake Hoffner <jhoffner@users.noreply.github.com>
Review on #2167: the ledger contract ends spec and epic rounds at human direction approval and records merge separately. Approval was re-read from each PR's timeline and the coordinator's in-session record. Six of the ten direction PRs record no approval act apart from the owner's merge. #1944 was approved on 89c1a1d and re-approved at merge after the owner's re-cut. #1957 was approved by label and closed unmerged, not merged as the first draft said. #2045 was approved at 17:10Z, three hours before its merge. No round count moved, because no review wave landed between an approval and its merge. Rounds-to-approval, design-off and the numbers the two proposals cite are unchanged. The fix A and B quotes now match the revised skill text, and the #2073 note records FIX-1564, filed at wrap. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016qaz1VpRK4wCQzxpjzZ4Hz


Two slots are still empty, and the third holds a ticket, not a result. The done condition is three proofs; one has a producer and none has been run. Kitchen-sink sits outside the box now — a sibling epic, not a member.
Why now. W3 made a Workforce describable and W4 made work reach a seat. Both ended in a claim about what exists. What stands between them and a launch is not more substrate — it is evidence, and evidence is what a QA epic produces.
The set: three exit proofs, one of them filed — FIX-1481, which owns Devtool checklist rows 4–6 while rows 1–3 ride FIX-1320 — plus the child that finished (FIX-1467,
references/vsresources/, merged — and never a proving leg), its two Backlog spin-offs FIX-1468 and FIX-1469, and FIX-1474, soft-related polish that is explicitly not an ER-Collab gate. As of 2026-09-21: 1 done · 2 backlog spin-offs · 1 backlog polish child · 1 filed proof-producer, not started · 2 named gaps with no child · 1 canceled. W5 has begun to staff its objective and has proved none of it. The set is open; filing against a proof is normal course, not a re-scope. Live table and dependency graph: SPEC.md.No lane carries a live bar — FIX-1481 is a ticket, not a duration. The ship fence stands — W4's first-cut issues are Done but FIX-1407 is In Review, so no W5 ship ticket opens; polish may start now.
Two empty cells on the diagonal: the done condition is owned in one part of three, and proved in none. Held is not passed — FIX-1481 is Backlog, and half of ER-Devtool rides an epic W5 does not run. ER-1, ER-2 and ER-3 were re-owned when kitchen-sink left, so no rule points at a row that is gone; ER-18 is marked MET.
Sign off
The objective gate is owed. Approving certifies the objective and the three exit proofs. The recalibration answered the previous three asks — they are not carried forward, and its own Still open items (exact checklist rows, which DevForce artifact counts, whether CyberForce gets a parallel thin proof) are explicitly not blocking EM start on polish, so none of them is an ask. The checklist rows are drafted as the EM's work.
One ask, answered in practice on 2026-09-20 — confirm or correct. Does W5 file its own Devtool child, or does ER-Devtool ride FIX-1320? FIX-1320 (Flow instances first-class, Spec Approved) already owns Devtool's instance and session surfaces; its own text says UI verification "is planned, not run." Three checklist rows read on exactly those surfaces. Recommendation: ride FIX-1320 for the instance/session rows and file a small W5 child for the workforce-specific ones — inventory, parked-row reasons, references vs resources, which are outside FIX-1320's stated scope. Changes my mind: FIX-1320 slipping past this cycle. If wrong: one Devtool pass done twice, or W5's proof hostage to another epic's schedule. What happened: FIX-1481 was filed taking exactly this recommendation — three rows to the W5 child, three riding FIX-1320, not the four-and-four the ask estimated. FIX-1320 is still Spec Approved with its UI verification unrun, so the changes my mind condition has not fired. Say so here if the split is wrong; nothing is built against it. Full six-part form in SPEC.md → Sign off.
News, not asks. Proof-via-DevForce is in and the invent-kill against it is dead (D8); kitchen-sink left the set (D9); ER-18 is met by FIX-1467; ER-15 was rewritten; D7 and D2 stand unchanged. FIX-1469's kitchen-sink half is stale — flagged for routing to FIX-1455, not moved. Two children joined the set table: FIX-1481 and FIX-1474, both already parented under FIX-1457 in Linear.
Reviewers · look here
references/document is indistinguishable from a mutable org resource. Both fail as written, so ER-Devtool is not a polish pass. Rows 1–2 are a source read, not a live run.Not here: any single issue's approach, the figures at pixel level, or the set table's status.
Spec · Decisions · Rules · Plan · Docs · Evolution · Linear FIX-1457 · Project: Workforce: Layer 2 Abstraction
This PR merges after the objective gate, and the set is retained at
specs/epics/FIX-1457/as reviewed design history. It no longer stays open for the life of the epic. After merge this body stays as the review record; material changes go on a follow-up PR frommain(orchestration.md → Merging and amending a spec).The recalibration, folded — what changed on Sep 20 at 18:05Z, and everything that moved with it
The authoritative source is FIX-1457's Linear body (rewritten 18:05:22Z); the PR comment
5751623639is its summary. This is the second re-scope today and the larger one: the first cancelled FIX-1458 and restated the objective as polish and prove on three surfaces; this one replaced the identity.The done condition, replaced rather than edited. ER-19 is retired. Three named exit proofs take its place:
The checklist is drafted, six pass/fail rows, because "looks good" is not a proof and the Architect left the detail to the EM. Each row says what passes and what the surface exposes today, read from
packages/devtool/srcandpackages/workforceonmain:task-collections-viewrenders counts, goal, status, assignee — but session-scoped; no standing board browserparkedis a status pill and a count; the reason is only inside the per-row JSON expanderFSDEV_DEBUG_ENDPOINTS=1; no org-level inventory viewEvery surface that restated the old answer, re-derived:
references/), the migration (clearShadowedReferences/describeShadowedReferences) and the grant model.Engineering calls — decided along the way, no product sign-off needed
SPEC.md2,276 against ~900,DECISIONS.md3,322 against ~1,300,BUSINESS-RULES.md2,400 against ~800,PLAN.md1,855 against ~900, and the two new documentsDOCS.md994 andEVOLUTION.md1,168. Every document is over. The rules document still carries ~450 words of six-row checklist the recalibration asked for, and the counts include table syntax. Reported rather than trimmed to fit.FIX-XXXplaceholder rows. The Devtool placeholder was replaced with the real id when FIX-1481 was filed, which is the convention working as intended. Inventing IDs for the other two, or leaving the work out because it has no ticket, remain the two worse options. Filing is the owner's.hireWorkforceregistering seats as flow instances; nothing below was observed against a running hired Workforce — which is the proof itself, and is why it needs a child.reviewRoundsis untouched — neither an owner recalibration nor a contract migration is review feedback.specs/epics/FIX-1457/, keyed by epic issue id, and gained the two documents the contract requires.node scripts/validate-spec-folder.mjsfailed withFound 1: spec/_epicsbefore the move and exits 0 after it. No decision, rule, invent-kill or scope changed in the move; the two set-table rows are the only content change, and every surface that restated zero producers was re-derived with them, figures included.Review round 1, folded — what four reviewers changed, and what they didn't
Head
1d9218ewas reviewed by greptile, cursor, codex andsecond-look. Folded intof65c529:How to review this — an epic-spec, not an implementation plan
This is an epic-spec: the shared objective and cross-cutting decisions for a set of
issues, in a retained document set. It is not any one issue's implementation design.
In scope to challenge:
That question can only be asked here.
DECISIONS.md— shared surface, naming, sequencing, contracts.BUSINESS-RULES.mdwith no owner, or two.DOCS.md; missing or misleadingpredecessor treatment in conditional
EVOLUTION.md.Out of scope — owned by the individual issue specs:
poc/show shape and scoping, not shippable implementation. Directional evidence, experimental
isolation and absence of secrets/generated dependencies remain in scope.
Feedback in the second list is routed to the issue it concerns as an implementer note, not
folded in here.
Required repository checks, approvals and review-thread policy still govern merge.
Optional comments are triaged, not a reason to grind to zero or restart the design review.
🤖 Generated with Claude Code
https://claude.ai/code/session_016qaz1VpRK4wCQzxpjzZ4Hz
Generated by Claude Code