Gate incoming pull requests on contribution evidence, not on guesses about who or what wrote them.
Standard library only - no runtime dependencies, no language model, no network calls other than the GitHub API.
For each incoming pull request PandaGate records observable facts:
- is an issue linked
- did tests change alongside logic
- how far the diff is spread across the tree
- does the description carry reproduction steps
- description substance relative to diff size
- author's prior pull requests in this repository
It then posts a single comment with those facts and the questions a maintainer would otherwise type by hand.
- It does not classify authorship. There is no "written by AI" verdict, flag, score or label anywhere in the output, and none is planned.
- It does not close, reject or request changes by default. The default
mode is
warn: one comment, no state change. - It does not check out or execute pull request code.
v0.0.1 - scaffold. No evidence rules are implemented yet. Do not read the sections above as a description of working behaviour.
Apache-2.0