conalh/CapabilityEcho
Code review for AI agent capability drift in pull requests.
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Stale
- License
- None
Inputs
| name | description | required | default |
|---|---|---|---|
| repo | Repository checkout path to inspect. Defaults to GITHUB_WORKSPACE. | no | "" |
| base | Base git ref or SHA to compare from. Defaults to the pull request base SHA when available. | no | "" |
| head | Head git ref or SHA to compare to. Defaults to the pull request head SHA when available. | no | "" |
| fail-on | Severity that fails the action. Use none, low, medium, high, or critical. Case-insensitive. | no | none |
| max-findings | If > 0, truncates the on-action markdown/json outputs and the step summary to the top-N findings ranked by severity. The full finding count, surface summary, rating, and fail-on decision are still computed against the complete set. 0 disables truncation. Use with report-file when you want both a short PR view and a full artifact. | no | 0 |
| max-output-bytes | If > 0, suppresses the report-markdown and report-json action outputs (only — not the step summary) when they exceed this byte size, replacing them with a short notice. 0 disables. Useful to stay inside GitHub Action output limits on very large PRs. | no | 0 |
| report-file | Optional path (absolute or repo-relative) to write the full markdown report. A sibling `<path>.json` is also written. Truncation never applies to the sidecar files, so pair this with `actions/upload-artifact` to keep a complete record alongside a truncated PR view. | no | "" |
| exceptions-file | Repo-relative JSON exception baseline to apply from the trusted base checkout/ref. Defaults to `.capabilityecho-exceptions.json` when present. Candidate-side exception changes are reported but do not take effect until merge. Entries use the shared agent-gov-core exception shape: kind, optional salientKey/pathPrefix, expires, and reason. | no | "" |
Outputs
| name | description |
|---|---|
| rating | Highest CapabilityEcho capability drift rating. |
| has-findings | Whether CapabilityEcho found at least one capability drift finding. |
| finding-count | Total CapabilityEcho findings in the diff. |
| changed-file-count | Number of changed scannable files in the diff. |
| analysis-incomplete | Whether CapabilityEcho skipped any changed input because it could not be safely examined. |
| analysis-diagnostic-count | Number of skipped-input or incomplete-analysis diagnostics. |
| analysis-diagnostics | JSON array of skipped-input or incomplete-analysis diagnostics. |
| suppressed-finding-count | Number of findings hidden by active checked-in exceptions. |
| expired-exception-count | Number of findings resurfaced because a matching exception expired. |
| surface-summary | JSON object with finding counts by executable surface. |
| severity-summary | JSON object with finding counts by severity. |
| capability-summary | JSON array of human-readable capability signal labels found in the diff. |
| top-recommendations | JSON array of the highest-priority recommendations for the diff. |
| adoption-evidence | Redacted JSON rollup for sharing in team feedback; excludes file paths and raw findings. |
| report-markdown | Full Markdown CapabilityEcho report, suitable for PR comments or downstream archival. |
| report-json | Full JSON CapabilityEcho report, suitable for local artifacts, dashboards, or downstream policy tooling. |