[kyle-sexton]

Refs: #3465

Summary

This PR delivers the fixture disposition for the provenance sweep sub-topic: the repository config now lets the attribution plugin's bundled test-fixture exclusion apply. The sweep that produced it, attribution-3465-close-20261106T023054Z, also covered the two other gaps in #3465's 2026-11-06 close audit. That evidence is in an issue comment (record), not in this diff.

Each #3465 checklist item and where it stands:

Item	State	Evidence
Golden set past min_n_per_class	met by owner ruling	near-verbatim n=15; the owner's 2026-11-02 ruling makes near-verbatim fix-eligible on its own (close audit, 2026-11-06)
Report-only sweep	done	the 2026-10-28 and 2026-11-04 sweep records on #3465
Live no-breadcrumb not-found probe	done	budgeted search on check-catalog.md:414-419: search 1 of 3 found the source, fetch 1 of 5 confirmed it (record)
Sweep resume semantics	done	an interrupted and resumed sweep restored spend 3, refused a re-close (exit 3), and re-validated the cache (same record)
Disposition-applying sweep	done	ran 2026-11-04 (applied 0); this sweep closed the fixture leave-with-reason, and the durable keep is this PR
Convention engagement, hash-store trigger	done	#13461
Dead-pointer round-trip	done	#13549 with melodic-software/ci-workflows#993
Narrow CI regression gate	deferred	owner decision Q-9759-2 (b), recorded by #13595

Fix

• docs/conventions/attribution.yaml: excluded_paths is removed. A layer's list replaces the bundled default whole (plugins/attribution/schemas/attribution.schema.json:12), so the repository's single **/evals/fixtures/** entry had silently dropped that default. The default also behaves differently from a copied list: it is matched below the scan target and skipped for a file named directly (list-corpus.sh:15-16, :305-313). The first cut copied its globs instead, and review showed that would decline a named fixture.
• The config comment says the eval harness's config isolation lifts the default, so eval fixtures still report their real findings (list-corpus.sh:17-19).

Verification

• list-corpus.sh on this branch: a repo-wide run declines 485 files under the bundled default reason, up from 277, so 208 more files in fixture and testdata trees are no longer audited as prose. scripts/fixtures/hooks-lifecycle-table.md named directly reads 1 included, 0 declined.
• The fixture itself is unchanged, and no per-file suppression was added.
• Left out: no eval run, because no skill or script changed.

Remaining

• The narrow CI regression gate for the cleaned state. Owner decision Q-9759-2 (b) defers it until a disposition sweep applies at least one disposition, and docs(upstream-drift): record the cleaned-state gate deferral and its trigger #13595 records the decision with its recheck trigger in docs/contracts/upstream-drift/README.md. Under the owner's 2026-11-05 direction, hardening-tied work closes and is rediscovered one item at a time. So provenance sweep sub-topic: repo-wide sweep, upstream-drift convention engagement, and the cleaned-state gate #3465 closes by hand once this lands, as upstream-drift: record the provenance convention entry and add the cleaned-state CI gate #9759 did.

Related

• provenance plugin: phases 6 to 8 (golden set and measurement, crosswalk registration, sweep) #3458 (golden-set growth)
• upstream-drift: record the provenance convention entry and add the cleaned-state CI gate #9759 (the gate's former tracker, closed under the same direction)

Follow-ups

• Pre-existing (advisory): plugins/machine-health/skills/audit/reference/windows/check-catalog.md:414-419 restates the Task Scheduler constants. The recommended disposition is condense-to-stamped-record, pointing at Microsoft's constants page, for a human to apply.
• Pre-existing (advisory): the attribution skill finds its config and sweep ledger from the session's working directory, so a run aimed at another worktree by path reads the wrong checkout's config.

🤖 Generated with Claude Code

https://claude.ai/code/session_01RMa441wb9nbFwyTS8nqdYj

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	⚠️ Failed 2026-11-06T02:37:48.597751Z We couldn’t complete this review. Comment @codex review to try again.	000a7b7	Draft marked ready

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[github-actions]

claude-security-review reviewed cb2ef0811d343a4767237d285e5ed2852e4adb25: read 1 of 1 changed file(s); 0 finding(s), 0 as inline comments.

Security review (incremental, blocking only): no security issues found

Scope: I read all of the change since b7e4bfb7 to the one listed file, docs/conventions/attribution.yaml. It matches none of the trusted-path globs. This lane already reviewed an earlier head (blocking only: yes), so this review reports only CRITICAL and IMPORTANT findings.

What changed: only the YAML header comment. It now says that the eval harness's config isolation lifts the bundled test-fixture default, so eval fixtures still report their real findings, and it cites the list-corpus.sh header. No config value changed: excluded_paths is still unset and trigger_less_stamp_check: true is unchanged.

What I checked:

• Behaviour: none changed. The change is comment text in a config file that only the attribution plugin's audit scripts, tests and evals read. No workflow, merge hold, kill switch, trusted-actor check or token broker reads it.
• Security impact: none. No secret is exposed, no token's scope grows, no new actor is trusted, no untrusted text reaches a lane that holds credentials, and nothing can bill the owner.
• Instruction surfaces: no standing instruction is deleted or weakened.
• Plugin manifests: no plugin.json or marketplace.json changed.
• PR context and settled threads: neither contains embedded instructions. The settled thread on this file (copied fixture globs) was fixed at b7e4bfb7, and this push doesn't reopen it.
• Prior findings: the earlier reviews by this lane found nothing, so nothing carries forward.

Spend: none.

critical-findings: 0

[github-actions]

claude-review: 1 inline finding(s) on 000a7b7523a5fe2f455970079eeea8601169c6fd; the summary is in this lane's pull request comment.

claude-lane: claude-review

[github-actions]

claude-review reviewed cb2ef0811d343a4767237d285e5ed2852e4adb25: read 1 of 1 changed file(s); 0 finding(s), 0 as inline comments.

Code review: PR #13643, changes since b7e4bfb7 (blocking only)

Result: no findings. This lane had already reviewed an earlier head, so this pass looked for blocking findings only.

What I read: the one file in scope, docs/conventions/attribution.yaml. I read every hunk of its change.

What changed: only the header comment. It gains one sentence saying that the eval harness's config isolation lifts the bundled test-fixture default, so eval fixtures still report their real findings. The sentence cites the header of list-corpus.sh. No YAML keys or values changed: excluded_paths is still unset and trigger_less_stamp_check: true is unchanged.

Is the comment accurate? Yes. The header it cites says so in two places:

• list-corpus.sh#L17-L19: "the harness lifts the default either way and the fixtures report real findings".
• L21-L27: the harness reads its config from a separate root, which is the isolation the new sentence refers to.

Earlier threads: the lane's earlier excluded_paths thread stays resolved. This push does not reopen it, because it changes no config.

Closing issue #3465: nothing changed since the last review. The close evidence is the author's record of sweep runs, which I did not verify.

Not applicable: no plugin manifest changed. I found no embedded instructions in the PR context, the settled threads or the prior findings.

[github-actions]

The pr-verify verifier failed head b7e4bfb731b4ce457b3a45283d7fed5f8b41ed92. What it wrote, as data:

What it wrote (28 lines)

Summary: The PR says `Closes #3465`, but its diff only removes `excluded_paths` from
  `docs/conventions/attribution.yaml`, so the bundled test-fixture default exclusion applies again.
  As a config change it looks sound. It is categorical, adds no per-instance suppression, and
  touches no secret, setting or takeover file. Every changed path is covered by a green job:
  lint-repo, check-plugins and test-bash (0) all passed
  (https://github.com/melodic-software/claude-code-plugins/actions/runs/42152259071/job/127679317535).

The problem is scope. #3465 is the sub-topic Brief, and its spec entry has no `person_checks`, no
  `credited_siblings` and no `scope`, so every checklist item counts in full. Several items are not
  delivered by this diff and are not excused for it. The issue's own 2026-11-06 close audit says the
  narrow CI regression gate for the cleaned state was moved to #9759, so nothing has delivered it.
  The convention engagement and hash-store trigger (#13461) and the dead-pointer wiring (#13549)
  shipped in other PRs, but the PR body has no `Siblings:` line, so they are not credited.
  Golden-set growth past `min_n_per_class` for verbatim (n=5) rests on an owner ruling, not on this
  change. The sweep-run items (budgeted search, interrupted-and-resumed sweep) are recorded in an
  issue comment, not in the diff. Closing the issue here would close it while checklist work is
  still open; the issue was already closed in error once before (#4960). Neither the PR body nor the
  diff contains an evidence-loop cap line or any text trying to steer the reviewer.
- PR body: change `Closes #3465` to `Refs: #3465`. Say that this PR delivers the fixture disposition
  (bundled fixture exclusion restored), and name what remains, at least the cleaned-state CI gate in
  #9759. If you keep `Closes`, add a `Siblings:` line naming the merged PRs that delivered the other
  checklist items (#13461, #13549), and make sure every remaining item, including the CI regression
  gate, is actually delivered first.
- docs/conventions/attribution.yaml:5: the deleted comment said the fixtures exclusion lived in
  config so the eval harness's config isolation would lift it. Confirm that the bundled default does
  not now hide `**/evals/fixtures/**` inputs during attribution eval runs (or that the
  scan-target-relative matching keeps them visible). Then state that in the new comment, so eval
  fixtures still report their real findings.

This pull request leaves the merge queue, and ci-status holds it out while it carries pr-verify-failed. A pr-verify pass removes the label, at a new commit or at this same head: once a trusted person edits the body, and only trusted people have edited it since the fail, this head is judged again, so a fail on the body or the issues it links needs only the body fixed, no new commit.