| name | self-driving-audit |
| description | Audit a repository for FullSelfDriving readiness: AGENTS.md quality, debug entry points, progressive discovery, observability, isolation, determinism, validation, safety, and evolutionary improvement readiness. |
Self-Driving Audit
Use this skill when the user asks whether a repository can be driven by an agent end-to-end, or asks how to improve a repo for agent operability.
Workflow
- Inspect the repo shape with fast local reads.
- Prefer root and scoped
AGENTS.md instructions as the first source of truth.
- Run the lightweight audit script when available:
python /path/to/fullselfdriving/scripts/fullselfdriving_audit.py <repo>
- Produce findings ordered by risk and leverage.
- Recommend the smallest repo-native upgrades.
Scoring Dimensions
- Agent instructions
- Debug entry point
- Progressive discovery
- Observability
- Isolation
- Determinism
- Validation
- Safety boundaries
- Evolution readiness
Output Shape
Use this concise structure:
Score: <0-100>, Level <0-5>
Highest-leverage gaps:
- ...
Recommended next changes:
- ...
Validation to add or run:
- ...
Rules
- Do not impose one debug framework across all repos.
- Do not mistake documentation for working capability.
- Prefer evidence that a command exists and produces artifacts.
- Call out destructive or production-adjacent workflows clearly.
