The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the joelclaw CLI to perform infrastructure management tasks. This includes modifying Redis state (clearing claims), re-firing events to Inngest, and potentially restarting Kubernetes deployments (system-bus-worker). These are high-privilege operations necessary for the skill's stated purpose of fixing stalled coding loops.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources which could be influenced by untrusted content.
Ingestion points: System state retrieved from Redis (PRD stories, progress entries, active claims), Inngest API health responses from localhost:3111, and local worktree content including .out files.
Boundary markers: Absent; the skill does not define specific delimiters to distinguish between diagnostic metadata and potentially malicious content within stories or plan entries.
Capability inventory: The skill can execute the joelclaw CLI to modify system state, restart services, and perform network requests via curl to local infrastructure endpoints.
Sanitization: No evidence of sanitization or validation is present for the system state data (e.g., story names or error logs in Redis) before it is processed by the agent during diagnosis.

loop-diagnosis