feat(hooks): refuse a CAT-UNVERIFIED tag from a turn that ran no check

[EdbertChan]

Summary

A reply that says "I could not check this" now has to show it tried.

If the reply tags a claim and the turn ran no check at all, the turn is refused
and must be written again.

The first slice only logged these claims. The reason given was that refusing
the reply would trap it in a loop.

That reason was wrong. Asking for an attempt is not asking for it to succeed.

Run the check; if it cannot settle the claim, the tag is then honest.

One release valve prevents the loop. A reply already being rewritten because of
this hook is let through, since that rewrite has no check of its own.

Review Claim

A well-formed unverified-claim tag in a turn that used no verification tool
fails that turn, unless the turn is already a rewrite prompted by this hook.

Review Lane

behavior

Review Unit

engine-runtime

Safety Invariant

The refusal needs three things at once: a well-formed tag in the reply, no
verification tool in that same turn, and no stop_hook_active. Any one of them
missing returns the previous allow, and the hook still touches only its own
ledger. The stop_hook_active release is covered by its own test, so the block
cannot become unescapable without that test failing.

Slice Rationale

The first slice landed the ledger and the parsing; this one changes only the
verdict. Split that way, a reviewer who dislikes the block can drop this commit
and keep the recording. Bundled, rejecting the block would also throw away the
ledger.

Non-goals

• Does not change diu-stop, hedge-runs-prove-it, or prove-it-ship-gate.
• Does not weaken any existing gate; every turn allowed before is still allowed.
• Does not ask whether the claim is true, only whether a check was attempted.
• Does not touch the reminder or the ledger format from the first slice.

Architecture

Before

graph TD
    A["reply carries a well-formed tag"] --> B["append to session ledger"]
    B --> C["turn ends, exit 0"]
    C --> D["reminder on the next prompt"]

Loading

After

graph TD
    A["reply carries a well-formed tag"] --> B["append to session ledger"]
    B --> C{"did this turn run Bash / Read / Grep / Glob?"}
    C -->|"yes"| D["turn ends, exit 0"]
    C -->|"no"| E{"stop_hook_active?"}
    E -->|"yes, a rewrite"| D
    E -->|"no"| F["refuse the turn, exit 2"]
    D --> G["reminder on the next prompt"]

Loading

Test Plan

Test Plan

• cd engine/hooks/unverified-tag-ledger && python3 -m unittest discover -s tests
• python3 engine/skills/make-pr/scripts/preflight.py --base origin/main
• python3 scripts/check_hook_test_coverage.py engine/hooks/unverified-tag-ledger

Replaying the real turn that motivated this, through the hook:

A) tag, zero tools      -> exit=2  "ran no verification tool. Untried claim(s):
                                    that it widened scope past the one session I gave it"
B) same tag after Bash  -> exit=0
C) stop_hook_active     -> exit=0  (no refusal loop)
D) no tag, no tools     -> exit=0

Ran 17 tests in 0.021s
OK
check_hook_test_coverage: OK (1 hook(s) checked)
ok      preflight passed

check_hook_test_coverage.py failed first, because this hook now blocks and had
no test whose name showed it firing. Renaming the test to
test_tag_with_no_attempt_is_blocked is what made it pass.

Fixture vs live: unit and fixture only. The hook is not installed on this
machine, so the refusal has never fired inside a real session:

$ ls -la ~/.claude/hooks/unverified-tag-ledger
ls: /Users/edbertchan/.claude/hooks/unverified-tag-ledger: No such file or directory
$ ls -d ~/.cache/catstack-unverified-ledger
ls: /Users/edbertchan/.cache/catstack-unverified-ledger: No such file or directory

Revert Plan

Revert Plan

• Safe to revert? Yes
• Revert command: git revert <sha> — leaves the ledger from the first slice in place
• Post-revert steps: None
• Data migration? No

[EdbertChan]

This pull request is part of a Mergify stack:

#	Pull Request	Link
1	feat(hooks): log CAT-UNVERIFIED tags instead of treating them as evidence	#464
2	feat(hooks): refuse a CAT-UNVERIFIED tag from a turn that ran no check	#468	👈
3	fix(hooks): wire unverified-tag-ledger into settings and declare its subagent opt-out	#474

[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_249e3121-361d-4dc4-bc1b-5303f0a7786f)