The Agent Skills Directory

[SAFE]: The skill is purely instructional and meta-analytical. It guides the agent through a process of reading and classifying its own system instructions (context) to improve efficiency.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process other instructions from files like CLAUDE.md and AGENTS.md. While these files could contain malicious instructions (indirect prompt injection), the skill includes a strong guardrail that requires the agent to only propose changes rather than apply them automatically. This ensures human-in-the-loop validation of all outputs.
[COMMAND_EXECUTION]: No shell commands or tool executions are defined within the skill; it relies on the agent's native ability to read and compare text files within its permitted environment.

context-audit