Repository navigation
Flag absence and capability claims as unverified-shaped in diu-stop - #1401
Merged
Merged
Conversation
'has no', 'there is no', 'no single', 'no such call' and 'can't be done' now trigger the unverified-claim check, so a no-such-call claim needs output or a tag like any other claim. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Change-Id: I6f50c350c18dfe9c42a78f4d5d17b48b151af2b9
Bugbot couldn't run - usage limit reachedBugbot 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: 007b7c44-e7f5-4ee5-8086-c39bb465d626) |
3 tasks done
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Replaces #1383: same change, rebased onto main. The original sits in a GitHub-managed PR chain that refused the merge and refused a base change, so this one is opened straight against main under the admin-bypass sweep.
Summary
A checker reads each reply before Claude sends it. It asks for proof whenever a claim sounds like a fact about the world.
The problem: claims like "there is no such call" or "that can't be done" slipped through with no proof, though they are claims too.
The fix: those phrases now count. A reply with one needs real output or an honest tag, like any other claim.
Replies without such phrases stay silent.
Review Claim
Claims that something does not exist or cannot be done now need proof or a tag before the reply is sent.
Review Lane
behavior
Review Unit
engine-runtime
Safety Invariant
Each change only affects when a checker warns about or blocks a reply. Checkers never edit files or run commands, and each change adds tests that fail before and pass after.
Slice Rationale
This slice stands alone: it only widens which phrases trigger the notice. The wording and retry changes sit in the previous slice, and the citation cleanup in the next one.
Non-goals
Measured
Command:
source scripts/test/hermetic_env.sh; python3 engine/hooks/diu-stop/tests/test_hooks.pywith this commit's fixtures and the parent's checker code (base) versus this commit's checker code (head).diu-stop: Ran 98 tests, FAILED (failures=5)diu-stop: Ran 98 tests, OKTest Plan
Test Plan
source scripts/test/hermetic_env.sh; python3 engine/hooks/diu-stop/tests/test_hooks.py-> Ran 98 tests, OKpython3 scripts/ci/check_hook_test_coverage.py engine/hooks/diu-stop-> check_hook_test_coverage: OK (1 hook(s) checked)Revert Plan
Revert Plan
git revert <merge commit of this PR>🤖 Generated with Claude Code
Note
Low Risk
Only widens Stop-hook phrase detection and adds fixtures; no changes to notice text, retry rules, or file/command side effects.
Overview
The diu-stop unverified-claim checker now treats absence and impossibility wording like other fact-shaped claims, so replies must show evidence or a
{{CAT-UNVERIFIED}}tag in the same paragraph.A new
ABSENCE_CLAIM_REpattern flags phrases such as has/have/had no, there is/are/was/were no, no single, no such call/api/endpoint/function/way, and can't/cannot be done._paragraph_claimruns this regex alongside the existing causal and hedge patterns. Fixture tests cover five firing cases and three silent cases (fenced output, well-formed tag, plain reply without those phrases).Reviewed by Cursor Bugbot for commit c00094b. Bugbot is set up for automated code reviews on this repo. Configure here.
🤖 Generated with Claude Code