release.yml restores the console dist from the same cache key but has no check:console-injection step — and its combined cache action saves even on failure

Filed by the PM seat from PR #9706's open question 2. Ruled A — this is the half PR #9706 could not take, because release.yml sits inside epic #9465's declared file territory.

What is missing

PR #9706 adds pnpm check:console-injection --require-stamp to ci.yml on both cache paths, so any run that consumes a restored console dist verifies that the dist actually carries this tree's @objectstack/spec rather than the published tarball. release.yml carries the same cache key and did not get the same step:

ci.yml:1439      key: ${{ runner.os }}-console-dist-${{ hashFiles('.objectui-sha', 'scripts/build-console.sh') }}
ci.yml:1485      (identical)
release.yml:491  (identical)   <-- unguarded

Why this is not covered by the ci.yml side

Two facts, both measured by #9706's dev, and the combination is what makes it reachable:

release.yml uses the COMBINED cache action, whose post-step saves even when the job failed — per ci.yml's own comment, which is why ci.yml deliberately uses a split restore/save instead. That is the concrete path a poisoned entry enters the shared, repo-scoped cache.
The console job is filter-gated. ci.yml's console paths filter is exactly .objectui-sha · scripts/build-console.sh · scripts/check-console-sha.mjs · .github/workflows/ci.yml — I verified this at ci.yml:102-106. So there is no guarantee any PR run inspects a given cache entry before release.yml restores it. A spec-only change never runs the console job at all.

⇒ "the ci.yml gate will catch it first" is not safe. It catches it first only when a PR happens to touch one of those four paths.

The change

Add, in release.yml after its check:console-sha step, the same step ci.yml now carries:

- name: Console injection stamp
  run: pnpm check:console-injection --require-stamp

Match ci.yml's spelling exactly rather than inventing a variant — the point is that both consumers of one cache key apply one check.

⛔ Fence

.github/workflows/release.yml is named explicitly in epic #9465's declared territory (.changeset/** · four changeset scripts · cut-rc.yml · release.yml · pr-automation.yml changeset steps · root package.json's @changesets/cli range and version script). It is not inside the scoping parenthetical, so it is fenced however the open question on that parenthetical resolves.

This card is for whoever holds the #9465 fence, or for after that epic completes. It is one step; the work is coordination, not implementation.

⚠️ Sequencing: blocked until PR #9706 lands — check:console-injection does not exist before it.

Refs: #9667 · PR #9706 · #8134 / PR #9660 · #9465

Blocked-by: #9465

Unlock scan (triage seat, 2026-08-19 ~09:1xZ round): the blocker was named only in prose (PR #9706), which merged 2026-08-19T00:04Z → pm:blocked removed; returns to the ungraded finding pool. Remaining constraint noted for grading/dispatch: release.yml sits inside epic #9465's declared file territory — coordinate with (or wait out) that epic rather than colliding with it. Filing hygiene: prose-only blockers are invisible to the unlock scan; use a Blocked-by: line.

Generated by Claude Code

First-touch grading (triage seat): promoted out of finding → pm:blocked, type Task.

Blocked-by: #9465

Premise re-checked on current values: PR #9706 merged 2026-08-19T00:04Z, so check:console-injection --require-stamp exists and the ci.yml side is guarded — the sequencing footnote is discharged. What remains is exactly the epic fence: .github/workflows/release.yml is named in epic #9465's declared territory (4/5 sub-issues complete as of this write). When #9465 closes, the standard unlock scan returns this card to the queue; alternatively the #9465 epic PM may absorb it as in-scope before closing — it is one step, and #9706's PR body already flags it as the natural completion. Re-check command for the eventual claimer: git grep -n "check:console-injection" origin/main -- .github/workflows/release.yml (zero hits = still needed).

Generated by Claude Code

feat(pm): H4/H14/blocking-index read comment-level Blocked-by lines (#…

feat(pm): H4/H14/blocking-index read comment-level Blocked-by lines (o…

Unlock scan: PR #9706 merged nine days ago — this card's stated blocker is discharged. It stays blocked, on the other one.

Triage seat, session session_01Aujz2zykf5LXt3T98gRsGe. pm:blocked retained — and this time that is applied, not just asserted.

The stated blocker is gone

The body reads "⚠️ Sequencing: blocked until PR #9706 lands — check:console-injection does not exist before it."

PR #9706 merged 2026-08-19T00:04:52Z (by os-steve, head 3497032ac). check:console-injection and its --require-stamp flag exist on main. That condition has been satisfied for nine days.

The card is still blocked, on the fence it names second

.github/workflows/release.yml is named explicitly in epic #9465's declared file territory, and #9465 is open — 4 of 5 sub-issues complete, still holding. PR #9706's own "Not touched, deliberately" section says the same thing from the other side: "Giving release.yml the same step is the natural completion and belongs to whoever holds that fence — flagging, not routing around it."

⇒ Both this card and the PR that created it agree the remaining constraint is the fence, not the tooling. So:

Blocked-by: #9465

That line is added here because the body carries none — which is the actual reason this needed a human-noticed catch rather than a scan.

⚠️ How late this is, and why

This card was not examined by the previous rounds' unlock scans at all. The scan enumerates open pm:blocked cards; it ran with a page size of 50 against a list of 51, and this was the card past the edge. It surfaced only because the count mismatch was noticed by hand this round.

⛔ Stating that plainly rather than quietly fixing it: nothing was lost — the discharged blocker did not unlock this card, because the fence is independent and still holds. But the scan reported completion over a list it had not finished reading, and that is the defect, not the outcome. Filed as #12817 along with the broader finding that 69% of pm:blocked cards carry no machine-readable Blocked-by: line anywhere in their body.

For whoever takes it when the fence lifts

Unchanged, and still one step. Match ci.yml's spelling exactly — pnpm check:console-injection --require-stamp — rather than inventing a variant; the whole point is that both consumers of one cache key apply one check. Re-verify the three key sites still agree at that time (ci.yml:1439, ci.yml:1485, release.yml:491 as measured), since nine days of main have moved under this card.

Generated by Claude Code

Unlock scan (R+89, triage seat, session session_019kDRpB7D2XzVzkaLp57T5D): upstream #9465 is CLOSED (completed). Card face re-verified on origin/main (a39b02a6): .github/workflows/release.yml contains no check:console-injection step (zero hits for console-injection; positive control: the file's actions/cache@v6 steps at :286 / :1042 / :1055 are reachable by the same grep), so a console dist restored from cache is still published unverified, and the combined cache action still saves on failure.

Released → pm:queue · priority:p2 · domain:devx · type Bug. p2 because the failure mode is a release artifact shipped with a stale or poisoned console dist and nothing red to say so. Lands in: release.yml — a check:console-injection step after the restore, and the cache save conditioned on success (or split restore/save so a failed build cannot seed the key).

Size/model suggestion: S–M, opus (workflow YAML is a place where a wrong if: is silent).

Generated by Claude Code

Claim: session session_01WLJQhde67SeTccsmnBVarV (domain:devx execution seat, seat post #6023) — R1 wave 7.

Branch: claude/issue-9709-release-console-injection-gate
Worktree: ../objectstack-9709 (dedicated per-task worktree, off origin/main)
Domain: domain:devx (CI/release workflow tooling; .github/workflows/** is NOT a governed surface)
File surface: .github/workflows/release.yml only — the console block at :1093–1110 on origin/main (79b6a22a5), re-verified just now: combined actions/cache@v6 (id: console-dist-cache, :1096) under the shared key ${{ runner.os }}-console-dist-${{ hashFiles('.objectui-sha', 'scripts/build-console.sh') }} (:1099), bash scripts/build-console.sh gated on cache-hit != 'true' (:1106–1107), pnpm check:console-sha (:1110), and zero hits for console-injection in the file (control: the same grep finds check:console-injection --require-stamp at ci.yml:1891 with the split restore :1822 / save :1898 shape around it). ⛔ ci.yml, scripts/check-console-injection.mjs and scripts/build-console.sh (held by build-console.sh strands a git worktree registration in the shared objectui checkout, and the next build dies on already registered #14177, PR fix(devx): prune stale objectui worktree registrations before build-console.sh adds one (#14177) #14417 landing) are NOT touched.
Container & model: subagent lane (S–M), opus — triage 5502993142 suggested S–M / opus ("workflow YAML is a place where a wrong if: is silent"); tier readout from scripts/pm/dispatch-gates.mjs --tier .github/workflows/release.yml: "no path-derived mandate — floor sonnet · default opus · ceiling fable".
Clause ②: not reached — a release-workflow gate step, no contract accept/reject behaviour, no public surface widened.
Serial constraints: epic Migrate the release toolchain to @changesets/cli v3 — one atomic PR carrying the bump, the pre-mode restructure, and the gates that model v2's semantics #9465 (the fence that held this card) is CLOSED; no in-flight devx card touches release.yml. scripts/build-console.sh is a cache-key INPUT, not an edit target here — build-console.sh strands a git worktree registration in the shared objectui checkout, and the next build dies on already registered #14177's edit to it busts the key once by design and does not interact with this card.

Generated by Claude Code

fix(ci): verify the restored Console dist's spec injection in release…

os-dev-report

{
  "issue": 9709,
  "status": "done",
  "branch": "claude/issue-9709-release-console-injection-gate",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/14424",
  "pr_number": 14424,
  "head_sha": "e405309146b5701f1a596795748b66282e4c6a2a",
  "premise_still_valid": true,
  "files_changed": [".github/workflows/release.yml (+34 -2, 1 file)"],
  "summary": "release.yml's publish job now applies the same two protections ci.yml applies around the shared console-dist cache key. Added 'Verify the restored Console dist bundles this tree's spec' (env CONSOLE_DIST_CACHE_KEY, run: pnpm check:console-injection --require-stamp) immediately after check:console-sha, and converted the combined actions/cache@v6 into the split actions/cache/restore@v6 -> build-on-miss -> assertions -> actions/cache/save@v6 (if: cache-hit != 'true') that ci.yml uses, so the combined action's save-even-on-failure post-step can no longer seed the shared key with a stamped-but-broken dist. Step id console-dist-cache kept (git grep finds its only two references inside this block). Scope held to release.yml: ci.yml, scripts/check-console-injection.mjs and scripts/build-console.sh were not touched.",
  "key_site_agreement": "PASS - 6 sites, one distinct string. git grep -h over ci.yml and release.yml piped through sort -u collapses every occurrence to exactly '${{ runner.os }}-console-dist-${{ hashFiles('.objectui-sha', 'scripts/build-console.sh') }}'. Sites today: ci.yml:1827 (restore key), ci.yml:1890 (CONSOLE_DIST_CACHE_KEY env), ci.yml:1903 (save key), release.yml:1106 / 1130 / 1142 (this PR). The card quoted ci.yml:1439 / ci.yml:1485 / release.yml:491 - main has moved far under it, and there are six sites now rather than three because both files also pass the key as an env var. Confirmed ci.yml and release.yml are the ONLY consumers of that cache path: git grep 'packages/console/dist' over .github/workflows finds no third cache; cut-rc.yml and showcase-smoke.yml build fresh, so assert-console-spec-injection.mjs runs inside build-console.sh there.",
  "tests": "No package test suite applies (workflow-only diff). Evidence: (1) YAML parse of release.yml with the workspace yaml@2.9.0 - OK, 4 jobs, publish has 19 steps. (2) Reachability walk over the parsed job, restore -> save window: [10] Restore (actions/cache/restore@v6, no if) / [11] build if cache-hit != 'true' / [12] pnpm check:console-sha / [13] pnpm check:console-injection --require-stamp with CONSOLE_DIST_CACHE_KEY / [14] Save (actions/cache/save@v6, if cache-hit != 'true'); 'steps inside the console window carrying continue-on-error / always() / failure(): 0', so the save is gated by Actions' default success(). Remaining combined actions/cache@v6 in publish: 2 - Setup pnpm cache, Setup Turbo cache (different caches, untouched). (3) Positive control that the added step has teeth AND that pnpm forwards the flag through a two-command script: 'pnpm check:console-injection --require-stamp' in this worktree (no packages/console/dist) -> exit 1, '. No console dist at packages/console/dist - nothing to verify, but a dist was required.', preceded by '. check-console-injection --self-test: 36 assertions'; the same checker without --require-stamp exits 0 on the same tree ('. No console dist ... skipping injection check'). (4) actionlint NOT AVAILABLE in this container (command -v actionlint -> not installed) - stated, not skipped silently. (5) Control-byte scan over the edited file: no hits. No ablation was performed - there is no built artifact in this diff to mutate; the equivalent evidence is (3), the measured red/green difference the added flag makes.",
  "gates": "21 of 21 derived families RAN-PASS, 0 NOT MEASURED, 0 UNRUN. Derived (not recalled) with 'node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands' from the worktree, letting the script take the change set from git itself. Machine reconciliation: 'node scripts/pm/dispatch-gates.mjs --ran ran.list' -> 'Run reconciliation - 21 derived, 21 run, 0 NOT-MEASURED, 0 UNRUN.' / '. dispatch-gates --ran: 21 derived famil(ies) accounted for - 21 run, 0 NOT-MEASURED.' (exit 0). Every exit code captured by redirect-then-read, never through a pipe. Verdict lines from the four families that actually read the edited file: check-step-collectors '390 run: steps across 30 workflow(s)'; check-workflow-status-functions 'OK (scanned 30 workflow file(s), 53 job(s), 26 job-level if: expression(s))'; check-required-contexts '6 required context name(s) pinned across 2 workflow(s)'; check:declared-population-live '156 of 200 famil(ies) declare a path population ... 7879 tracked file(s)'. Standing devx extras also run: check:ratchet-remedy-authority PASS ('183 scripts swept'), check:declared-population-live PASS (also inside the 21). NOTE on check:pm-dispatch-gates: its first run was cap-killed at the container's ~10 min foreground ceiling (exit 143, no verdict). That is NOT MEASURED, not a pass - and parseRunRecord's own header records a card where exactly that kill was written off as NOT-MEASURED and came back green. So it was re-run to completion rather than claimed: os-verify-lock slot issue-9709, waited 7m35s, held 12m57s, command-exit 0, '. dispatch-gates self-test: 1232 cases pass.' Hence 0 NOT-MEASURED in the record above. The always-runs tail (33 unconditional CI steps this derivation names no family for, including pnpm lint) is outside the derived total and is CI's, per the standing division.",
  "changeset": "NONE - skip-changeset label applied to PR 14424 and read back (additive POST /issues/14424/labels -> HTTP 200; independent GET read-back -> ['skip-changeset']). Justification from the gate's own text, not precedent alone: pr-automation.yml's changeset-check job states its exemption as 'this PR declares no release of its own', and lint.yml names 'this PR edits a CI-internal script' as the textbook case. This diff is one workflow file and publishes nothing from any package. There is no path-based auto-exemption for .github/** - the label is the only mechanism.",
  "mcp_calls": "0 - every GitHub read and write went through REST (issue body, 5 comments, PR create, label POST, label read-back, PR body read-back). No MCP GitHub tool was called.",
  "open_questions": [],
  "out_of_scope_findings": [],
  "observations": [
    "The card body's snapshot of the step to copy ('- name: Console injection stamp' / 'run: pnpm check:console-injection --require-stamp') predates ci.yml's current shape. 'Match ci.yml's spelling exactly' was read as matching what ci.yml carries TODAY: step name 'Verify the restored Console dist bundles this tree's spec', the run line, and the env CONSOLE_DIST_CACHE_KEY. That env is not cargo-cult - check-console-injection.mjs:818 reads it and remedy(cacheKey) uses it so a refusal names the real key an operator has to evict, instead of printing a template.",
    "ci.yml also carries an 'Assert a real Console dist is in place' step before check:console-sha, which this PR deliberately does NOT copy. Under --require-stamp a missing dist and an unstamped dist are already fatal (verified live: exit 1 on a tree with no dist; the checker's self-test pins 'no dist fails when required' and 'unstamped is fatal when required'), so the assert's three checks would be redundant here. That step exists in ci.yml to protect its BARE-invocation reasoning, not the --require-stamp path. Recorded rather than acted on.",
    "The save step sits right after the console assertions rather than at the end of the publish job - matching ci.yml. Deliberate: the entry vouches for the console dist, and every question this repo asks of one is green by then; deferring to the end of a job that ends in npm publish would mean a run that failed to publish also failed to bank a console build it had fully verified.",
    "origin/main moved from b2eab9594 to a69dfde80 during this run; no upstream commit touches .github/workflows/release.yml since the branch point (git log b2eab9594..origin/main -- .github/workflows/release.yml is empty), so the branch is conflict-free on its one file."
  ]
}

Generated by Claude Code

ACCEPT — #9709 landed. PR #14424 merged to main as cd36b8420 via the merge queue (auto-merge armed after the CI watcher read GREEN; the auto_merge: null / mergeable_state: unknown reading while queued is the PR mid-queue, not disarmed).

Close-out probes executed on origin/main (fresh fetch, throwaway worktree — not the shared checkout; cd36b8420 confirmed as an ancestor of origin/main at a98b61b3e):

.github/workflows/release.yml:1131 run: pnpm check:console-injection --require-stamp — the release workflow now invokes the console-injection gate in stamp-required mode, which is the card's deliverable.
Control: the same check:console-injection pattern reads 2 hits in .github/workflows/ci.yml, so the grep is live and the release-workflow hit is the diff.

pm:dispatched stripped in this same step (type / priority / domain:devx kept — ownership is not state).

Seat: session_01WLJQhde67SeTccsmnBVarV (domain:devx, #6023).

Generated by Claude Code

Navigation Menu

Repository navigation

release.yml restores the console dist from the same cache key but has no `check:console-injection` step — and its combined cache action saves even on failure #9709

What is missing

Why this is not covered by the ci.yml side

The change

⛔ Fence

Unlock scan: PR #9706 merged nine days ago — this card's stated blocker is discharged. It stays blocked, on the other one.

The stated blocker is gone

The card is still blocked, on the fence it names second

⚠️ How late this is, and why

For whoever takes it when the fence lifts

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

release.yml restores the console dist from the same cache key but has no check:console-injection step — and its combined cache action saves even on failure #9709

Description

What is missing

Why this is not covered by the ci.yml side

The change

⛔ Fence

Activity

os-warren commented on Aug 19, 2026

os-warren commented on Aug 19, 2026

huangyiirene commented on Aug 27, 2026

Unlock scan: PR #9706 merged nine days ago — this card's stated blocker is discharged. It stays blocked, on the other one.

The stated blocker is gone

The card is still blocked, on the fence it names second

⚠️ How late this is, and why

For whoever takes it when the fence lifts

huangyiirene commented on Sep 2, 2026

claude commented on Sep 2, 2026

claude commented on Sep 2, 2026

claude commented on Sep 2, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

release.yml restores the console dist from the same cache key but has no `check:console-injection` step — and its combined cache action saves even on failure #9709