The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to run local Python scripts for structural code analysis. Evidence: Instructions to use python .claude/scripts/code_graph query callers_of <function> --json.
[PROMPT_INJECTION]: Potential surface for Indirect Prompt Injection through processed files. Ingestion points: docs/project-reference/domain-entities-reference.md and codebase files. Boundary markers: Explicit instruction to check for [Injected: ...] headers. Capability inventory: Subprocess execution via local Python scripts and file writing to plans/reports/. Sanitization: No specific argument sanitization described for graph queries.
[SAFE]: The skill enforces mandatory evidence-based reasoning protocols, requiring file:line proof and minimum confidence thresholds (80%) for all recommendations, which effectively mitigates the risk of hallucinated or unverified optimization suggestions.

performance