Repository navigation
Add repeatable approval-boundary regression checks - #18
Merged
Merged
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Add repeatable regression checks for the eight approval scenarios, including fail-closed grading and evidence that the installed skill actually loaded. Normal CI runs only offline tests; live Codex and Claude Code trials are opt-in.
Validation: 72 offline tests passed locally. Prior local live trials against installed v1.1.1 returned 8/8 structured decisions on each host with loader evidence. These are single samples, not a reliability estimate or proof of safe execution.
The README preserves two manual reasoning cautions: Claude case B suggested a broader approval bundle than the proposed PR required; case E overstated generic duplicate-push consequences. Automated PASS grades four decision fields and does not certify free-text reasoning.
No skill version, installed source, release workflow, tag, or published release changes. Raw host event logs remain private and are not committed.