[CAT-UNVERIFIED tag](3) The engine rules tell authors to write the tag

[EdbertChan]

Summary

engine/CLAUDE.core.md now tells an agent to run the check first and use the tag only when the check cannot run.

It still said to write bare UNVERIFIED:, which the hooks now reject. An agent following it would be blocked for obeying it.

The rewrite also changes the order. The old rule made the marker the default answer to missing evidence. The new rule puts the check first.

Review Claim

Approve the engine rules naming {{CAT-UNVERIFIED: <claim> -- cannot verify: <reason>}} instead of bare UNVERIFIED:, plus a helper that finds lines still instructing the old marker.

Review Lane

behavior

Review Unit

engine-runtime

Safety Invariant

Prose plus one read-only scan helper; no hook, install or runtime path changes.

Slice Rationale

Rules come after the behavior they describe. Corpus skills, the docs files and the product skill are separate review units, so they are the next three PRs.

Non-goals

• No hook or install change.
• Lines where the word means something else are left alone: "UNVERIFIED schema" (Codex hooks.json) and "UNVERIFIED end-to-end" (a status cell). So are recorded transcripts and fixtures.

Test Plan

Test Plan

• python3 engine/skills/make-pr/scripts/preflight.py --base <this PR's base>

unit    engine-runtime: 4 file(s)
neutral 1 file(s): tests/escape_hatch_vocab.py
declare Review Unit: engine-runtime
gate    python3 scripts/check_codify_has_code.py --base stack/EdbertChan/cat-unverified-tag/2-retry-checks-evidence
        ok      codify-has-code
gate    python3 scripts/check_no_dated_provenance.py --base stack/EdbertChan/cat-unverified-tag/2-retry-checks-evidence
        ok      no dated provenance
gate    python3 scripts/check_hook_test_coverage.py engine/hooks/wrong-check-reflect
        check_hook_test_coverage: OK (1 hook(s) checked)
gate    python3 scripts/check_skills_three_harnesses.py
        ok      skills three-harness check
gate    python3 scripts/check_ecosystem_boundaries.py
        ok      ecosystem boundaries
gate    python3 scripts/check_skill_file_refs.py
        ok      skill file refs
gate    python3 scripts/check_skill_test_coverage.py --base stack/EdbertChan/cat-unverified-tag/2-retry-checks-evidence --head HEAD
        ok      skill test coverage
gate    python3 scripts/check_skill_trigger_mechanism.py
        ok      skill trigger mechanism
gate    python3 scripts/check_skill_trigger_policy.py
        ok	skill trigger policy
gate    python3 scripts/check_subagent_scope_contract.py
        ok	subagent scope contract
gate    python3 scripts/run_skill_scenarios.py
        ok  	subagent-fact-attributed
        ok  	stale-green-caught-without-any-admission
        ok  	stale-green-already-owned-by-the-reply
        ok  	red-then-green-is-a-fix-not-a-stale-claim
        
        ok	all 19 scenario(s) behaved as declared
ok      preflight passed

• bash scripts/run_all_tests.sh -- exit 0, 45 suites OK, 0 failed, 1676 tests
• tests/escape_hatch_vocab.py flags 30 instructing lines on 509b9cd and 0 with the whole stack applied
• python3 -m unittest discover -s engine/skills/make-pr/tests

Revert Plan

Revert Plan

• Safe to revert? Yes
• Revert command: git revert <sha>
• Post-revert steps: None
• Data migration? No

🤖 Generated with Claude Code

https://claude.ai/code/session_01VKsvxJk65w6q7KnPSRYvNg

---

Note

Low Risk
Documentation and a vocabulary-scan helper only; no hook, install, or runtime behavior changes in this slice.

Overview
Engine prose and the make-pr fixture-vs-live gate now match what evidence hooks enforce: run the check first, and only when verification truly cannot run, mark claims with {{CAT-UNVERIFIED: <claim> -- cannot verify: <reason>}}. Bare UNVERIFIED: is explicitly retired in engine/CLAUDE.core.md, evidence-check.mdc, and make-pr/SKILL.md so agents are not told to use a marker hooks reject.

A new read-only helper tests/escape_hatch_vocab.py scans instructional files for lines that still tell authors to write bare UNVERIFIED: (ignoring “retired” mentions and non-marker phrases like “UNVERIFIED schema”). make-pr adds unit tests that the skill names the CAT tag and passes that scan.

^{Reviewed by Cursor Bugbot for commit 6a541b6. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_35c1ed16-a96c-49dd-86ba-039bac5e7c6a)

[EdbertChan]

Mergify repair stopped: lowest open stack PR #417 is based on stack/EdbertChan/cat-unverified-tag/2-retry-checks-evidence, which still belongs to open PR(s) #416 outside this stack; leaving it alone to avoid dropping dependency changes

[mergify]

Tick the box to add this pull request to the merge queue (same as @mergifyio queue).

• Queue this pull request

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_c274087d-4c83-421b-9da3-b0c06ad587df)