-
Notifications
You must be signed in to change notification settings - Fork 1
Description
Problem
The PR title/body fingerprint check is producing false positives when normal prose mentions AI tooling names (for example, "Cursor" or "Copilot").
Observed run:
- Workflow run: https://github.com/vig-os/devcontainer/actions/runs/23002702709
- Job: https://github.com/vig-os/devcontainer/actions/runs/23002702709/job/66791079620?pr=270
Note: that specific job failed first on PR title format, but the same workflow includes check-pr-agent-fingerprints, and local repro confirms the false-positive behavior.
Reproduction
Run locally:
PR_TITLE='docs: mention copilot integration' \\
PR_BODY='This PR updates Cursor docs with plain text mention of Copilot as comparison.' \\
uv run check-pr-agent-fingerprintsCurrent result:
- exits non-zero
- prints:
PR title or body contains blocked AI agent fingerprint: 'cursor'
Expected behavior
Plain-text references to tools/vendors in documentation or explanatory text should not fail the check by default. The check should block agent identity fingerprints (trailers, bot identities, explicit attribution patterns), not generic mentions.
Likely root cause
.github/agent-blocklist.toml includes broad name substrings under [patterns].names, including:
cursorcopilot
Because matching is case-insensitive substring-based, any ordinary text mentioning those words is rejected.
Suggested direction
Narrow matching scope so it targets identity attribution patterns instead of broad substrings. For example:
- keep strict trailer checks (
Co-authored-by, etc.) - restrict name matching to attribution contexts (author lines/signatures), or
- add explicit allow patterns/contexts for normal prose references.