The Agent Skills Directory

[COMMAND_EXECUTION]: The file evals/run-static-checks.sh is a shell script intended for developer-side static analysis. It invokes local Python scripts to validate skill structure, representing a standard maintenance tool.
[PROMPT_INJECTION]: The skill processes untrusted agent traces and tool outputs as part of its 'optimization-loop' and 'autonomous-controller' workflows. This creates an indirect prompt injection surface. Ingestion points: Spans and trajectories containing external model interactions (references/playbooks/optimization-loop.md). Capability inventory: Creation of audit reports and ledgers (templates/audit-report.md), and staging of automated improvements (references/playbooks/autonomous-controller.md). Sanitization: The playbooks mandate privacy filtering and human review of all changes. Boundary markers: Explicit delimiters for untrusted trace data were not identified in the prompt templates.

agent-ops