fix(outcome): attribute capture to the exercised skill, not brigade-work

[solomonneas]

Summary

• Decision for outcome: generic brigade-work artifact id collapses distinct skill signals #628: hard-coded brigade-work is a defect. Distinct skills should accumulate their own signals; brigade-work remains the fallback when no artifact is known.
• Verify receipts stamp outcome capture intent. Retry guidance uses the failed receipt stamp before the current --capture value.
• Claude hooks record a skill only when its id is installed in the target. Reads of reference skills outside the target do not affect attribution.
• Automatic card attribution is intentionally out of scope for this PR. Explicit card captures remain supported; automatic attribution is limited to installed target skills.
• Existing append-only outcome rows are unchanged.

Closes #628.

Verify receipt

20260801-200756-work-verify-fce43d

brigade work verify run --target . --command "./scripts/verify" --capture brigade-work

Acceptance criteria to tests

Criterion	Test
Distinct skills accumulate separate rank signals	test_distinct_skill_captures_accumulate_separate_rank_signals
Retry uses the failed receipt stamp over a current capture	test_capture_before_retry_prefers_receipt_stamp_over_current_capture
Retry uses a receipt-stamped artifact	test_capture_before_retry_uses_receipt_stamped_artifact
Retry uses current --capture when the receipt is unstamped	test_capture_before_retry_uses_current_capture_when_receipt_unstamped
Unknown skill falls back to brigade-work	test_capture_before_retry_falls_back_to_brigade_work
External SKILL.md reads do not set attribution	test_posttooluse_read_external_skill_does_not_set_exercised_artifact
Installed SKILL.md reads set attribution	test_posttooluse_read_skill_sets_exercised_artifact

Test plan

• Full project verification
• Rebased onto origin/main
• External grader review before merge

[coderabbitai]

Important

Review skipped

Auto reviews are limited based on label configuration.

🏷️ Required labels (at least one) (1)

• coderabbit-review

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository: escoffier-labs/coderabbit/.coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: d5436bbb-ccbd-4d53-9e58-4a52a23a8950

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[solomonneas]

Grader review: APPROVE-WITH-CHANGES. Attribution mechanism and backward compatibility are sound; ledger rows untouched. Required before merge:

1. Rebase: branch is CONFLICTING with main - fix(outcome): dedupe reused verify receipts via reused_from #650 rewrote _verify_receipt_has_outcome_capture in the same verification.py region.
2. Priority inversion in _enforce_capture_before_retry: resolve_capture_artifact_id puts the CURRENT run's --capture ahead of the failed receipt's own stamp, so retrying skill B's verify after skill A's failed run cross-attributes A's hurt signal to B. The receipt stamp is ground truth for that run; swap the order and add the both-present conflict test (existing tests only cover one-source cases).
3. Constrain the last-SKILL.md-read heuristic: the path matcher accepts any */skills//SKILL.md anywhere on disk, including other repos and meta-skills browsed for reference. Constrain to artifacts installed in the target (capture already has _artifact_known) or explicitly document last-read-wins and filter meta-skills.

Also state in the PR body whether card attribution (issue #628 says 'skills or cards') is intentionally out of scope. Ping for re-grade after push.

[solomonneas]

Rebased onto origin/main in 6fce03d.

a561695 maps the grader changes as follows:

• Retry attribution now prioritizes the failed receipt outcome_capture artifact id over the current --capture value. Covered by test_capture_before_retry_prefers_receipt_stamp_over_current_capture.
• Read-path attribution now requires an installed target skill. Covered by test_posttooluse_read_external_skill_does_not_set_exercised_artifact.

Card attribution scope is stated in the PR body: automatic card attribution is out of scope; explicit card capture remains supported.

Full verification passed: ./scripts/verify, receipt 20260801-200756-work-verify-fce43d.

[solomonneas]

Grader: APPROVE, merging. All three required changes verified: the branch is rebased (6fce03d), the priority inversion is corrected so the failed receipt's own outcome_capture stamp wins over the current --capture value (test_capture_before_retry_prefers_receipt_stamp_over_current_capture covers the both-present conflict that the earlier tests missed), and the read-path heuristic now requires an installed target skill (test_posttooluse_read_external_skill_does_not_set_exercised_artifact). Card attribution scope is stated explicitly in the PR body rather than left ambiguous. Receipt 20260801-200756-work-verify-fce43d, CI green.