abhinavag-svg/AI Autonomy Risk Tier Evaluator
Deterministic, no-LLM pre-merge risk classifier for AI-authored code changes. Scores each pull request against a versioned policy (rubric + CODEOWNERS + optional service map), posts a PR comment + Check Run with the resulting risk tier (0-3), and supports an auditable human /override command. Pure-function scoring -- no model is in the decision path.
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Stale
- License
- None
Inputs
| name | description | required | default |
|---|---|---|---|
| policy-path | Path to policy.yaml file (relative to repo root) | no | policy.yaml |
| codeowners-path | Path to CODEOWNERS file (relative to repo root) | no | .github/CODEOWNERS |
| service-map-path | Path to a standalone service-map YAML file (same shape as policy.yaml rubric.service_map: service_name -> [dependent_service, ...]). When set, overrides any rubric.service_map embedded in policy.yaml. | no | — |
| enable-drift-check | Enable deterministic control/architecture drift gate and PR comment | no | false |
| control-catalog-path | Optional controls.yaml mapping for drift impact candidates | no | — |
| drift-policy-path | Optional future drift policy path; currently reserved for compatibility | no | — |
| fail-on-drift-severity | Fail control-drift-gate on unacknowledged drift at this severity or higher: none, low, medium, high | no | high |
| drift-comment-mode | Drift PR comment mode: summary or off | no | summary |
| github-token | GitHub API token for comment posting | yes | — |
Outputs
| name | description |
|---|---|
| decision-id | Unique decision record ID |
| risk-score | Numeric risk score (0-5) |
| tier | Authorization tier (0, 1, 2, 3) |
| triggered-rules | JSON array of triggered rubric rules |
| enforcement-level | Human-readable enforcement action |
| drift-count | Number of drift flags detected |
| unacknowledged-drift-count | Number of open drift flags without acknowledgment |
| max-drift-severity | Maximum drift severity: none, low, medium, high |
| control-impact-count | Number of control impact candidates |
| evidence-gap-count | Number of drift/evidence gaps |