openclaw-self-eval

An agent setup audit checks foundational state including soul files, memory stores, manifests, skills, and persona to surface misconfigurations and drift. It validates presence and alignment, then flags gaps with actionable fixes in a readable report.

A soul file audit verifies the soul file exists, is non-empty, and has been recently updated. This ensures the agent's foundational configuration maintains freshness and prevents operational staleness before issues impact performance.

Yes, vault and secrets audits resolve references without exposing actual secret values. All checks run in read-only mode, validating configuration security while generating reports that maintain strict confidentiality of sensitive credentials.

Verifying agent manifest and skill registry alignment confirms AGENTS.md lists the agent and matches the persona. It ensures each registered skill directory contains SKILL.md and skill.yml files with proper registry alignment.

Reporting configuration drift involves comparing the current agent state against a last snapshot. The audit generates a human-readable Markdown report plus a machine-readable JSON sidecar, detailing deviations for quick actionable resolution.

No, the self-eval audit runs entirely in read-only mode. It checks reachability, freshness, and alignment of memory stores, installers, and persona without modifying files, emitting results as a Markdown report and JSON sidecar.