The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs legitimate file system operations using Bash and Python to list, read, and write instinct and cluster files within the ~/.claude/instincts/ directory. This usage is transparent and tied directly to the skill's stated purpose.
[PROMPT_INJECTION]: The skill processes data from local instinct files to generate summaries and 'skill hypotheses'. While this involves ingesting external data into the agent's context (Indirect Prompt Injection surface), the skill is designed to target only verified ('active' or 'proven') content from the agent's own history, which significantly mitigates risk.

evolve