storkme/PR Second Opinion
Independent, agentic, OpenRouter-backed second-opinion review on a PR.
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Stale
- License
- None
Inputs
| name | description | required | default |
|---|---|---|---|
| openrouter-api-key | OpenRouter API key (https://openrouter.ai/keys). | yes | — |
| github-token | Token with pull-requests:write. Default: the workflow token. | no | ${{ github.token }} |
| pr-number | PR number to review. Default: the PR from the triggering event. | no | ${{ github.event.pull_request.number }} |
| model | OpenRouter model id for the review passes. | no | z-ai/glm-5.2 |
| k | Agentic passes to union. K=1 (default) skips the merge. | no | 1 |
| merge-model | OpenRouter model for the K>1 merge. Default: same as model. | no | "" |
| project | Project name used in the review prompt. | no | ${{ github.event.repository.name }} |
| guidance-file | Path (in the repo) to this project's review checklist / 'memory'. | no | "" |
| exclude-globs | Comma-separated globs to drop from the diff. Default: sensible set. | no | "" |
| max-diff-chars | Cap on the diff size sent to the model. | no | 60000 |
| pass-timeout-seconds | Per-pass timeout. | no | 900 |
| tools | Agent tool grant. 'read,bash' (default) greps for callers/tests; set 'read' to drop shell access on repos with untrusted PR authors. See README Security. | no | read,bash |
| reasoning | Whether the model is a reasoning model. Default true (the default z-ai/glm-5.2 is). Set false for a non-reasoning model so pi omits reasoning params. | no | true |
| fail-on-degraded | Fail the check (exit 2) when a pass degrades — times out, exits non-zero, or produces no review — and no review was posted. Surfaces reviewer malfunction, not review findings (a posted review always passes). Set 'false' for the old always-green behavior. | no | true |
Outputs
no outputs