The Agent Skills Directory

[SAFE]: No security vulnerabilities or malicious patterns were detected. The skill follows best practices for AI agent safety, including:
Read-only Constraints: The instructions explicitly forbid the agent from running commands that change or record entities, focusing purely on retrospective auditing.
Human-in-the-loop Adjudication: The operating model mandates that all generated reports are advisory and require human review and action, preventing autonomous system modifications.
Security Guidelines for Sources: The external source contract in references/source-contract.md explicitly prohibits reading credential directories or exfiltrating data to undeclared services.
Data Boundaries: The system uses node-specific claims (nodeId and claimFence) and offline replay to prevent unauthorized shared state, pretend fleet-wide snapshots, or self-editing loops.

harness-reckoning