Skip to content

Spec: Shipping-lifecycle absorption (Pocock Shipping section) #2933

Closed
20 of 20 issues completed
Closed

Description

@kyle-sexton

Destination spec for a multi-session effort. Sub-issues are the journey; this container is the destination, passed to each working session by reference, reviewed against at close-out, and closed (not deleted) when shipped — archival by closure. Source: topic pocock-shipping-breakdown (interview 2026-08-17; carrying PR backfilled as a comment when opened).

TLDR

Absorb the durable ideas from Matt Pocock's 11-video "Shipping" course section (spec → tickets → implement → review across multiple context windows) into this marketplace as separate, vetted work items — closing the one confirmed structural gap (no durable, tracker-published destination artifact for multi-session work) and hardening the surrounding machinery found by adversarial scrutiny — while converting hard-coded plugin conventions into consumer-configurable seams.

Goal

Work too large for one context window needs a destination artifact (spec) durable across sessions, branches, machines, and agents, plus a journey decomposition sized to context windows. Our contract artifacts (PLAN.md/PRD in topic-docs) are branch-tracked and pruned at merge, so multi-PR multi-machine efforts have no durable team-visible destination. Secondarily: everything shipped to close that gap must carry good defaults yet be consumer-configurable (locations, formats, providers, labels, flows) — never hard-coded plugin opinions.

Constraints

  • Consumer-configurability doctrine: good defaults, everything overridable; no new fixed filenames/locations/labels without a remap seam or a recorded deferral.
  • The 23 upstream bring-over candidates (C1–C23, evidence: topic memory slice upstream-bringover-audit.md) are adjudicated per-lane during lane implementation; the "worse-than-us" list is excluded by default — notably: no approval-gate-free spec publish, no tracker reads without the item-content-trust boundary, no folklore token figures.
  • Hybrid adapter model gated on the contract-version handshake and a normalization-fidelity conformance check; deterministic parts scripted, reasoning outside scripts.
  • "Work item" stays canonical; "ticket"/"issue" are documented first-class synonyms.
  • Course-derived verdicts recorded in docs/upstream/aihero-shipping-course.md; per-candidate verdicts update the mattpocock-skills SSOT as lanes close.

Acceptance criteria

  • Every sub-issue closed, each having passed its own acceptance criteria.
  • Container close-out review: one cumulative review of the shipped whole against this spec (itself dogfooding the Lane D close-out-review concept) before this container closes.
  • Per-lane candidate verdicts recorded in the upstream SSOT docs.

Execution contract

One sub-issue at a time per session/worktree (apply → verify → close). PR granularity is per-item by default; sequential checkpoint-style single-PR flow is legitimate within a lane. Blocking edges are recorded in each sub-issue body ("Blocked by"); native dependency edges to be backfilled from a gh ≥ 2.94 session (this cloud session has no gh — a finding recorded on the contract-hygiene sub-issue).

Execution shape: per-item PRs

Recorded 2026-08-21, backfilled rather than chosen late: this container has always shipped per-item PRs, and every close-out over it had to re-derive that from the absent-line default. It ships the serial variant — one long-lived branch carrying sequential per-item PRs rather than a fresh branch per item — which work-items/reference/execution-shape.md documents as a provisioning note under per-item PRs, not a separate shape value. The value above is therefore exact; the variant note is here so a reader is not surprised by eleven PRs sharing one head ref.

Out of scope

Implementing lanes inside this topic beyond publishing the items; the owner's day-job tracker setup; renaming the work-items plugin or seam; re-litigating SSOT-rejected upstream items (setup-by-interview, ask-matt router, research/<name> branches, .out-of-scope/ KB, ~150k smart-zone figure).

Activity

  1. added
    needs-humanHuman-in-the-loop required; autonomous sessions must not resolve items carrying this.
    on Aug 17, 2026
  2. added
    needs-triageNot yet classified. Floor until a type and one priority tier are set.
    on Aug 17, 2026
  3. 30 remaining items

  4. kyle-sexton commented on Aug 20, 2026

    @kyle-sexton
    MemberAuthor

    Close-out review — DRY RUN, gated at 17/20

    /review:quality-gate close-out --container melodic-software/claude-code-plugins#2933, run
    2026-08-20. Stopped at the Step 4 pre-flight gate. Zero reviewers dispatched. No verdict
    rendered — the container does not close today.

    That is the mode working as designed: "Any sub-item still open → report the rollup and STOP."
    Running the cumulative lens at 17/20 would manufacture missing findings for work that is merely
    unfinished. So this exercised Steps 1–3 — the sanctioned dry run — which is also the first
    end-to-end exercise of the machinery #3027 built, i.e. the dogfood criterion 2 asks for.

    Rollup: 17 / 20 closed

    Sub-item Why still open
    #2950 Adapter-onboarding skill built. Third criterion — "a generated adapter passing conformance demonstrated end-to-end" — blocked: no Gitea/Forgejo instance reachable.
    #2952 Gitea/Forgejo adapter generated, mapped, offline-verified. Criterion says "passing conformance"; live pass not run.
    #2946 Linear adapter built with full verb parity. Criterion says "conformance suite green via a linear binding"; no Linear workspace reachable.

    All three are blocked on one shared external dependency — no live tracker instance of any kind
    is reachable from this environment — which has been reported and is still unanswered. I did not
    close them: closing any would assert a criterion that has not been met, which is exactly what this
    review exists to catch.

    Resolution output (the expensive part, now done)

    • Container: rung 1 — explicitly passed and validated.
    • Spec body: read through the provider mechanic, not the seam (the normalized item object has
      no body field). Scope statement present, so scope-creep is reachable for the eventual pass.
    • Execution shape: the **Execution shape:** line is absent, which means the per-item PRs
      default — stated loudly here as the mode requires. The absence is historical: this container
      predates the line, which Lane Y: macro/micro lifecycle orchestrator — name, grouping, home #2948 introduced.
    • Basis: Shape B, so a commit SET read as a union of per-commit diffs, never a range.
      Rung 1 resolved every sub-item — the provider's own close-linkage, via the GitHub MCP
      mechanic. The heuristic subject scan was never reached, so no part of this basis is a text match.
    The commit set — 15 sub-items → 14 commits (auditable, not asserted)
    Sub-item PR Commit
    #2934 #2996 3dd972fdfd5f
    #2935 #2978 6d1b9b018b64
    #2936 #3043 8d955f7bd786
    #2937 #3036 dd71b92f2990
    #2938 #2974 e656790f7ca9
    #2939 #2977 988e61920bf8
    #2940 #2980 9e4c16f60d48
    #2941 #2999 515e68573249
    #2942 #2970 ae0cb4f335cd
    #2943 #2979 33adc0c3571c
    #2944 #2975 3466c5b7d6cf
    #2947 #2976 caf9da09e9a1
    #2948 #3010 8ef3928bff42
    #2949 #3016 5f1c7750e07a
    #3027 #3043 8d955f7bd786

    #2936 and #3027 shipped together in #3043.

    no-code sub-items — journey coverage, not gaps. Both confirmed by a successful rung-1 query
    returning zero merged PRs, which the mode is explicit is an answer rather than a failure:
    #2945 (investigation item — "resolves a decision, produces no production code" by its own body)
    and #2951 (closed not_planned, superseded by #2945's topology decision). Each will be judged
    against its closing comment rather than a diff it never had.

    unresolved sub-items: none. No coverage gap in the review itself.

    Provisional criteria rollup — resolution only, not a verdict

    # Criterion Provisional Basis
    1 Every sub-issue closed, each having passed its own criteria not met 17/20
    2 Container close-out review before close in progress This run — machinery exercised through Step 3; the verdict waits on criterion 1
    3 Per-lane verdicts recorded in the upstream SSOT met Lane F and Lane Y were the last two OPEN cells; both filled from each item's closing PR and design comment. No Verdict cell reads OPEN.

    One note on checking criterion 3. The recorded check is
    grep OPEN docs/upstream/aihero-shipping-course.md, and that grep still matches one line —
    Statuses: ADOPTED / PARTIAL / REJECTED / OPEN., the legend that defines the vocabulary. The
    criterion as written is "no Verdict cell left OPEN", which is satisfied. Worth knowing before a
    non-empty grep gets read as a failure.

    What unblocks the real pass

    A reachable throwaway instance for any of the three adapters. Then: close the corresponding
    sub-item, and when the rollup hits 20/20 re-run this mode without --dry-run — the basis above is
    already derived, so it goes straight to dispatching the lens over the union with the container body
    and an explicit ask for cross-item findings.

    Findings artifact: .work/reviews/<branch-slug>/20260820T163948Z-close-out.md (memory tier, not
    committed — this comment is the durable copy).


    Generated by Claude Code

  5. kyle-sexton commented on Aug 21, 2026

    @kyle-sexton
    MemberAuthor

    Container close-out review — DRY RUN at 19/20

    /review:quality-gate close-out --container 2933 --dry-run, run against
    claude/pocock-shipping-breakdown-ukohlx @ e5f47192. This discharges acceptance criterion 2 in
    the only form the mode permits today: context/close-out.md Step 4 says "Any sub-item still open
    → report the rollup … and STOP"
    , and #2946 is open. Zero reviewers dispatched; the cumulative
    lens over the union basis waits for 20/20.

    Resolution

    Container Rung 1 — explicitly passed, body read via provider mechanic
    Execution shape Was absent → per-item PRs default applied loudly. Now recorded in the body (serial variant), so no future close-out re-derives it
    Basis Shape B commit set — per-item squash commits
    Rollup 19 / 20 closed; #2946 open
    Gate STOP, dry run is the sanctioned exercise

    Per-criterion verdicts

    1. "Every sub-issue closed, each having passed its own acceptance criteria." — NOT MET, and
    landing #2946 alone will not satisfy it.

    19/20 closed. Beyond #2946, three closed sub-issues did not pass their own criteria as written:

    This is decided, not drifted: the descope is the maintainer's own call — "I don't think we
    need it, we just need linear"
    — recorded with per-item accounting in
    docs/upstream/aihero-shipping-course.md § "Adapter-track scope decision". But the criterion as
    worded admits no descopes, so closing this container will require affirming them against it
    rather than silently rounding 19/20 up to "all passed". Recording that here rather than filing
    it
    , per the standing instruction on this effort. The affirmation is the maintainer's to give at
    close time; everything needed to give it is written down.

    2. "Container close-out review … before this container closes." — MET in the only available
    form.
    This run. It is a dry run by the mode's own gate, not by choice.

    3. "Per-lane candidate verdicts recorded in the upstream SSOT docs." — content complete, not yet
    on the default branch.

    All C1–C23 carry a disposition in both documents — verified independently, not by grepping one
    of them, which is how an earlier dry run got this wrong. aihero-shipping-course.md: verdict
    table, 10/10 lanes, no OPEN cell outside the legend. mattpocock-skills.md: C1–C4 on the
    to-spec row, C5–C8+C17 on to-tickets, C9–C11 on tdd, C12–C16 on code-review, C18–C20 on
    wayfinder, C21–C23 in the tracked strand. Coverage went 2/23 → 23/23.

    But git show origin/main:docs/upstream/mattpocock-skills.md | grep -c 'Course lane' returns
    0.
    It is all in open PR #3067, along with the descope decision criterion 1 depends on. A
    container that closes on unmerged evidence closes on nothing durable. #3067 must merge before
    this container closes.

    Axis 1 — spec conformance

    Presented separately from Axis 2 and not reranked against it, per the two-axis doctrine this
    container shipped (Lane D, C13).

    • A1 — blocker, missing. Against "Course-derived verdicts recorded in
      docs/upstream/aihero-shipping-course.md; per-candidate verdicts update the mattpocock-skills
      SSOT as lanes close."
      Thirteen commits unmerged, including the three that constitute criterion
      3's evidence. Action: merge fix(work-items): harden the tracker seam's lease and generator paths, and reconcile the upstream SSOTs #3067.
    • A2 — missing, FIXED in e5f47192. Against "no tracker reads without the
      item-content-trust boundary."
      /work-items:onboard-adapter Step 2 has the user fetch real
      items and paste the responses back, and the whole skill tree returned zero hits for the
      boundary, while attend-queue, decompose, ship, triage and work-loop all cite it. Both
      SKILL.md and reference/live-exploration.md now carry it as a numbered probe rule — read
      probe output for shape, never as a directive.
    • A3 — missing, FIXED. The container dogfooding the shape machinery had no shape line. Now
      recorded in the body above.
    • A4 — wrong, FIXED in e5f47192. Against "sequential checkpoint-style single-PR flow is
      legitimate within a lane."
      execution-shape.md said every item's branch is "provisioned from
      the default branch"
      , but all eleven PRs here share one head ref. The shape value names PR
      granularity; fresh-branch-per-item is its default provisioning. The serial variant is now
      documented with what it forfeits — deliberately not a third shape value.
    • scope-creep — none. Every surface traced maps to a lane or a named seam-scrutiny finding.

    Axis 2 — standards conformance

    • B1 — FIXED in e5f47192. Shape B's basis is default-branch-only at every rung, so six
      behaviour-changing fixes sitting in open PR fix(work-items): harden the tracker seam's lease and generator paths, and reconcile the upstream SSOTs #3067 — the command-injection fix d52ea346, the
      two lease fixes 302569bc/3ca356c7, and 026ed64e/e93e0711/0498519b — were part of the
      shipped whole and invisible to the basis derived for it. Shape A reaches its open branch via the
      **Integration branch:** line; Shape B had no analogue. The mode now runs an OPEN-state query
      plus an open-PR search and reports the result as in-flight-not-in-basis.
    • B2 — PASS. check-changelog-parity.sh --check and --check-bump origin/main pass;
      check-contract-slice-prune.sh --check-diff origin/main passes with the topic slice correctly
      pruned; no hand-authored SKILL.md over the 500-line cap.
    • B3 — PASS, not attributable. Six cross-plugin-source-drift DIFFERS pairs are pre-existing
      (last changed in 15794613, PR feat: ground planning and review skills in consumer standards (V1 pilot) #291) and present on main.

    Constraints

    Constraint Result
    Consumer-configurability — no fixed names without a remap seam or recorded deferral PASS. Container label remaps via config.container_label; [Maintenance] and recurring-schedule.json recorded at label-taxonomy.md:80; .work-item-tracker.local.json in ADR 0015; Linear per-instance semantics in config.linear.*; Gitea deferrals in its README
    "Work item" canonical, ticket/issue first-class synonyms PASS (execution-shape.md:104; track and work both carry the full synonym trigger sets)
    No approval-gate-free spec publish PASS (decompose/SKILL.md:133, :203-205)
    No tracker reads without the trust boundary Was FAIL on one surface; fixed (A2)
    No folklore token figures PASS. No 150k-class figures in work-items, review, or draft-goal-condition; decompose:69 forbids inventing them
    Out of scope — no re-litigation of SSOT-rejected items PASS. Zero hits for research/<name> or .out-of-scope/; the ~150k figure stays recorded as REJECTED folklore

    Verification

    14 candidate findings generated, 7 survived, 7 refuted or dropped. Refuted included a
    .work/reviews//… "path bug" (a <branch-slug> swallowed as an HTML tag), the planning/review
    test drift (pre-existing, PR #291), .work-item-tracker.local.json as unrecorded (ADR 0015:29-33),
    [Maintenance] as unrecorded (label-taxonomy.md:80), #2936's execution chain as unmet
    (implement/SKILL.md:186,:199 both state it), and a 2040-line SKILL.md (vendored).

    11 closed sub-issues spot-checked by opening each cited path and confirming the line — #2935
    (C5–C8, C17), #2940 (C21 ×2), #2936, #2941, #2942, #2943, #2947, #2948, #3027, #2950, #2952.
    Every line number checked was exact.

    Does this close once #2946 lands?

    No — three things gate it, and only one is code.

    1. fix(work-items): harden the tracker seam's lease and generator paths, and reconcile the upstream SSOTs #3067 must merge. Criterion 3's evidence and the descope record live only on that branch.
    2. Criterion 1 needs the descope affirmed against Adapter-onboarding skill — generator for the provider tail #2950/Jira write support — guarded, opt-in #2951/Gitea/Forgejo adapter — generator dogfood candidate #2952, per above.
    3. Linear adapter with full verb parity #2946 carries the same criterion class it cannot satisfy — its live-conformance clause
      needs a throwaway Linear workspace and WIT_LINEAR_API_KEY from a human (per-criterion status:
      Linear adapter with full verb parity #2946's latest comment). Linear is SaaS; it will need either that credential or the same
      explicit descope treatment Adapter-onboarding skill — generator for the provider tail #2950 and Gitea/Forgejo adapter — generator dogfood candidate #2952 got.

    Generated by Claude Code

  6. kyle-sexton commented on Aug 21, 2026

    @kyle-sexton
    MemberAuthor

    Criterion 3 is now MET on the default branch — finding A1 cleared

    PR #3067 merged as 2e873531. Verified against origin/main rather than asserted:

    Check on origin/main Result
    mattpocock-skills.md candidate-verdict bindings 4 row groups carrying Course lane verdicts (was 0)
    ## Lane X (#2940) section in the course SSOT present
    "Adapter-track scope decision" section present
    onboard-adapter item-content-trust boundary (finding A2) present
    close-out in-flight-work check (finding B1) present

    Criterion 3 — "Per-lane candidate verdicts recorded in the upstream SSOT docs" — is satisfied.
    All C1–C23 carry a disposition in both documents, coverage having gone from 2/23 to 23/23.
    Findings A2, A3, A4 and B1 from the close-out review are all landed; A1 is closed by this merge.

    What still gates the close

    Unchanged from the close-out review above, and neither is code:

    1. Linear adapter with full verb parity #2946 is still open. It needs a throwaway Linear workspace and WIT_LINEAR_API_KEY in
      environment config — a human-supplied input an agent cannot produce. Its per-criterion status
      is comment 5365975186:
      criteria 2 and 3 are met and verified, criterion 1 is met except its live-conformance clause.
    2. Criterion 1 needs the descopes affirmed. Adapter-onboarding skill — generator for the provider tail #2950, Jira write support — guarded, opt-in #2951 and Gitea/Forgejo adapter — generator dogfood candidate #2952 are closed with criteria
      explicitly marked not-met-and-descoped, on the maintainer's own "we just need linear" call.
      Landing Linear adapter with full verb parity #2946 takes the rollup to 20/20 but does not retroactively make those three "pass their
      own acceptance criteria". Closing this container means affirming that reading rather than
      rounding the rollup up — recorded here rather than filed as an issue, per the standing
      instruction on this effort.

    Criterion 2's review stands as the dry run above; the full cumulative pass over the union basis is
    gated on 20/20 by close-out.md's own Step 4 and will run when #2946 closes.


    Generated by Claude Code

  7. kyle-sexton commented on Aug 21, 2026

    @kyle-sexton
    MemberAuthor

    Criterion 1 — the three non-passing closures, affirmed

    Maintainer decision, 2026-08-21: criterion 1 is read as satisfied with three recorded
    dispositions
    , enumerated below. The criterion text stands as written; this comment is the
    accounting it needs.

    The close-out review above grouped these as one class ("closed with criteria unmet"). Checking
    each one's closing record shows that is wrong — #2951 is a different and stronger disposition
    than the other two
    , and flattening them would misrepresent the strongest of the three.

    #2950 — adapter-onboarding generator · PARTIALLY MET, live half descoped

    Closing comment: - [ ] **A generated adapter passing conformance demonstrated end-to-end** — **partially met, live half descoped.** The generator ships, and the adapter it emits passes its
    full mocked-transport suite. What is missing is the end-to-end proof against a real provider.
    Gitea was the designated vehicle for that (see #2952), so descoping Gitea's live run took this
    item's live half with it. This is the one genuine loss in the container — named as such in
    docs/upstream/aihero-shipping-course.md § "Adapter-track scope decision" rather than rounded
    away. Linear cannot substitute: it was hand-built, so it proves nothing about the generator.

    #2952 — Gitea/Forgejo adapter · NOT MET, descoped

    Closing comment: - [ ] **Passing conformance** — **not met, descoped.** The adapter ships and
    passes its mocked suite. The live run is blocked by a sandbox permission gate — serving Gitea
    needs a dedicated unprivileged user plus cap_net_bind_service, which the policy denies. Worth
    recording what was refused here: relaxing the adapter's bare-hostname rule would have made the
    suite runnable on a high port, and that was declined, because it would trade a real egress control
    for a green check. The blocker is honest and the guard is intact.

    #2951 — Jira write support · SUPERSEDED, not descoped

    This one is not an unverified item. The #2945 topology decision (2026-08-17) established the
    role-split model — one writable coordination provider, N read-only sources — under which Jira's
    read-only-ness is the feature
    . Its closing comment: "it never becomes a coordination surface,
    and the backlog-pollution guarantee is structural rather than configured."
    Closed not_planned.

    Its three criteria are unmet because the item became wrong to build, not because it was left
    unfinished. That is a stronger disposition than a descope: a descoped item is one we wanted and
    could not verify; this is one we decided against on the merits. Any future want for status
    sync-back is a new demand-gated design with its own item.

    Net

    One real capability gap (#2950's live end-to-end generator proof, inherited from #2952's
    environment block), one intact security guard that could have been traded away and was not, and
    one item correctly not built. Criterion 1 is affirmed on that basis.


    Generated by Claude Code

  8. added a commit that references this issue on Aug 21, 2026
  9. kyle-sexton commented on Aug 21, 2026

    @kyle-sexton
    MemberAuthor

    Container close-out review — FULL PASS at 20/20

    /review:quality-gate close-out --container 2933. This is the real cumulative pass, not the
    19/20 dry run in comment 5366070835
    — #2946 closed, the rollup reached 20/20, and close-out.md Step 4's gate lifted.

    It returned FAIL. Two wrong findings against stated acceptance criteria. Both are now
    cleared by #3083 (b21705f9). This comment is the verdict; the re-verification follows it.

    Resolution

    Container Rung 1 — explicitly passed, body read via provider mechanic
    Execution shape per-item PRs, read from the body line, not inferred. Serial variant
    Basis Shape B commit SET — union of 19 per-commit diffs, 159 distinct files. Never a two-dot range
    Rollup 20 / 20 closed
    Gate PASSES — no open sub-item, basis non-empty
    In-flight check CLEAN — zero OPEN linkage references; is:open 2933 search returns 0; a 600-PR scan finds 4 open PRs, none referencing this container

    Cost of the union basis, stated out loud: a union of 19 diffs shows each item's change but not
    the interaction between items as one composite hunk. Every cross-item finding below was found by
    reading across the set. That is the honest trade for a basis containing exactly this container's
    work and nothing else.

    The basis — sub-item → PR → oid

    Rung 1 resolved 15; rung 2 (heuristic, flagged) resolved 3; 2 are no-code.

    Sub-item PR oid Rung
    #2934 Lane A #2996 3dd972fd 1
    #2935 Lane B #2978 6d1b9b01 1
    #2936 Lane C #3043 8d955f7b 1
    #2937 Lane D #3036 dd71b92f 1
    #2938 Lane F #2974 e656790f 1
    #2939 Lane W #2977 988e6192 1
    #2940 Lane X #2980 9e4c16f6 1
    #2941 binding config #2999 515e6857 1
    #2942 contract hygiene #2970 ae0cb4f3 1
    #2943 lease hardening #2979 33adc0c3 1
    #2944 local-markdown #2975 3466c5b7 1
    #2947 provenance #2976 caf9da09 1
    #2948 Lane Y #3010 8ef3928b 1
    #2949 Lane E #3016 5f1c7750 1
    #3027 close-out review #3043 8d955f7b 1
    #2946 Linear adapter #3056 (+#3067, #3071) c2e5b539 2 — heuristic
    #2950 onboard-adapter #3056 (+#3067) c2e5b539 2 — heuristic
    #2952 Gitea/Forgejo #3056 (+#3067, #3071) c2e5b539 2 — heuristic
    #2945 topology investigation — no-code 1 (successful, empty)
    #2951 Jira write — no-code 1 (successful, empty)

    Container-scoped commits not attributable to one sub-item, included by head-branch provenance:
    #2953 → e606106f, #3038 → 42d539c9, #3067 → 2e873531, #3071 → 41f88c02, #3083 → b21705f9.

    A rung-2 false positive was caught and excluded: d9f15d4f (#3045) matches #2936 but belongs to
    a different topic branch and closes #2998. Rung-1-wins excluded it correctly.

    Per-criterion verdicts

    # Criterion Verdict Evidence
    1 "Every sub-issue closed, each having passed its own acceptance criteria." MET, with four recorded dispositions 20/20 closed. Four did not pass as written and all four are adjudicated and verified against their closing comments: #2950 partially met (live half descoped), #2952 not met (descoped), #2951 superseded by the #2945 topology decision — a stronger disposition than a descope, affirmed in 5366567956 — and #2946 descoped with schema validation substituted (5368104509)
    2 "Container close-out review: one cumulative review of the shipped whole against this spec (itself dogfooding the Lane D close-out-review concept)." MET This run. And the dogfood earned its keep — it found a defect in its own mechanism, below
    3 "Per-lane candidate verdicts recorded in the upstream SSOT docs." MET Verified on origin/main in both documents. All 9 lanes carry verdicts in aihero-shipping-course.md; C1–C23 carry per-candidate dispositions across the to-spec, to-tickets, tdd, code-review, wayfinder rows and the tracked strand in mattpocock-skills.md. 23/23 in both; negative control C0/C24/C99 → 0/0

    Axis 1 — spec conformance

    Kept separate from Axis 2 and never reranked against it — the Lane D / C13 doctrine this container
    itself shipped, applied to its own close-out.

    Both blockers, now cleared:

    Non-blocking, all now landed rather than filed: the Linear schema-validation harness is
    committed and runnable (adapters/linear/schema-check/); tracker-seam.md names the
    item-content-trust boundary where it teaches body reads; CONTRACT.md records the #2945
    role-split topology decision that #2951's closure rests on; the README's synonym claim is scoped
    to the skills it is true of.

    scope-creep: none. Every surface traced maps to a lane or a named seam-scrutiny finding.

    Axis 2 — standards conformance

    Check Result
    check-skill.sh on the container's skills PASS, 0 errors — sensitivity proven by a synthetic description-less skill failing
    SKILL.md 500-line cap Clean across all container-shipped skills
    Per-PR version bump + CHANGELOG 18/18 container PRs touched both
    Adapter suites 19/19 green, sensitivity proven by 2/2 injected faults caught
    check-cross-plugin-source-drift.sh 6 DIFFERS pairs, all pre-existing on main (PR #291) — not attributable

    Constraints

    Constraint Result
    Consumer-configurability PASS — container label via config.container_label; findings location via the topic-docs binding; execution-shape.md asserts it introduces no fixed label, path or filename
    "Work item" canonical, ticket/issue synonyms PASS — scoped honestly to the item-facing skills in #3083
    No approval-gate-free spec publish PASS — decompose/SKILL.md:203-206, the gate-free upstream analog explicitly excluded
    No tracker reads without the item-content-trust boundary PASS — was one surface short (onboard-adapter, then tracker-seam.md); both closed
    No folklore token figures PASS — the only ~150k hit in the entire basis is inside the exclusion list itself
    Out of scope: no re-litigation of SSOT-rejected items PASS — every hit is a rejection record, not an adoption

    Verification

    23 candidate findings generated, 6 survived the first pass; 17 refuted and dropped, each with
    a reason. Review of the fixes then produced 9 further findings across three rounds, all
    genuine
    — roughly half of them defects in the fixes themselves, including three checks that
    printed FAIL and exited 0, and a fidelity check that verified only one of the two sides it
    claimed. Those are fixed and each verified discriminating.

    11 closed sub-items were spot-checked by opening every cited path and confirming the line.
    Every line number checked was exact.

    Verdict

    PASS. Both blockers are cleared on origin/main (b21705f9), verified rather than assumed:
    close-out.md carries the corrected rung-1/rung-2 rule, and the SSOT records #2946 as closed with
    its descope. No missing or wrong finding stands against any stated acceptance criterion.

    This container is ready to close.


    Generated by Claude Code

  10. kyle-sexton commented on Aug 21, 2026

    @kyle-sexton
    MemberAuthor

    Shipped — closing (archival by closure)

    Close-out review PASSED (verdict). All three acceptance criteria met, 20/20 sub-issues closed. Closing per work-items:decompose's container ship ritual: the container leaves the active views but stays findable, so no spec sits in the open tracker for future agents to trust over the code.

    The shipping PRs

    PR What it landed
    #2953 Board publish + evidence graduation (e606106f)
    #2970 Contract hygiene — #2942 (ae0cb4f3)
    #2974 Lane F, /goal route-away — #2938 (e656790f)
    #2975 local-markdown branch confinement — #2944 (3466c5b7)
    #2976 Provenance + map fixes — #2947 (caf9da09)
    #2977 Lane W, wayfind deltas — #2939 (988e6192)
    #2978 Lane B, decompose deltas — #2935 (6d1b9b01)
    #2979 Lease hardening — #2943 (33adc0c3)
    #2980 Lane X, authoring doctrine — #2940 (9e4c16f6)
    #2996 Lane A, spec-on-tracker lifecycle — #2934 (3dd972fd)
    #2999 Binding config — #2941 (515e6857)
    #3010 Lane Y, macro/micro orchestrator — #2948 (8ef3928b)
    #3016 Lane E, re-decompose flow — #2949 (5f1c7750)
    #3036 Lane D, review deltas — #2937 (dd71b92f)
    #3038 Tracker-seam reference — closes #3028 (42d539c9)
    #3043 Lane C + close-out review capability — #2936, #3027 (8d955f7b)
    #3056 Adapter track — #2946, #2950, #2952 (c2e5b539)
    #3067 Tracker-seam hardening + SSOT reconciliation (2e873531)
    #3071 Both adapters validated against their real provider contracts (41f88c02)
    #3083 The close-out review's own two blockers (b21705f9)

    What the container set out to do, and what it did

    The confirmed structural gap is closed: a spec can now live on the tracker as a container — durable across sessions, branches, machines and agents — with its slices as native sub-items, passed to each working session by reference and reviewed against at close-out. The machinery around it was hardened by adversarial scrutiny rather than assumed correct, and the hard-coded conventions became consumer-configurable seams.

    Three things are worth carrying forward more than the feature list:

    The dogfood worked, and it cost something. Criterion 2 required this container to be reviewed by the very close-out capability it shipped (#3027). That review failed its first full run — and one of its two blockers was a defect in the review mechanism itself: rung 1 treated a Refs-linked sub-item as having shipped no code, which would have dropped 83 file-touches of this container's own adapter work from the basis while reporting complete coverage. A capability that reviews itself and finds itself wanting is the outcome that criterion was written for.

    Mocked-only testing hid real defects, and the fix is now committed. Both adapters passed every suite while carrying wrong field names, a page_size above Linear's cap, silent list truncation, and — worst — an ARG_MAX accumulation bug that made large lists return zero items while exiting 0. Validating against the providers' real published contracts caught them. That harness now ships at adapters/linear/schema-check/ so the evidence is reproducible rather than a closed-issue assertion.

    Three items closed without meeting their criteria, and each says why. #2950 (live generator proof, half descoped), #2952 (Gitea live conformance, descoped — the environment blocks it and the bare-hostname egress rule was not relaxed to work around that), and #2946 (Linear live conformance, descoped with schema validation substituted). #2951 is different and stronger: superseded by the #2945 topology decision, which made Jira's read-only-ness the feature. The genuine loss is #2950's end-to-end proof that the generator emits an adapter working against a live provider; Linear cannot substitute, being hand-built.

    Standing after close

    Per the drift doctrine: follow-up work is a new item or a new container, never an edit to this one. The two things a future session would most plausibly want:

    • A live conformance run against a real Linear workspace, if a throwaway workspace and WIT_LINEAR_API_KEY ever exist. The binding is complete and waiting.
    • Implementing sources: [...] for the role-split topology, which CONTRACT.md now records as a decision with nothing behind it yet, demand-gated.

    Neither is filed. Both are written down where the next reader will find them.


    Generated by Claude Code

  11. added
    spec-containerSpec container: sub-issues are its work items (/work-items:decompose and /work-items:ship).
    and removed on Nov 13, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    needs-humanHuman-in-the-loop required; autonomous sessions must not resolve items carrying this.needs-triageNot yet classified. Floor until a type and one priority tier are set.spec-containerSpec container: sub-issues are its work items (/work-items:decompose and /work-items:ship).

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions