The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local shell scripts (scan-skills.sh and scan-rules.sh) to index the user's skills and rules directories. These scripts use standard, non-destructive system utilities like find, grep, and awk to perform read-only discovery of configuration files.
[INDIRECT_PROMPT_INJECTION]: The skill processes text from external skills which could theoretically contain malicious instructions. However, the risk of these instructions influencing the system is mitigated by the skill's design, which requires principles to appear in at least two different sources and mandates human review for all proposed changes.
Ingestion points: Reads the content of SKILL.md and rule files from local directories (~/.claude/skills and ~/.claude/rules).
Boundary markers: The analysis prompt utilizes Markdown headers (## Input) to clearly separate the ingested skill data from the agent's instructions.
Capability inventory: Performs file system discovery and content reading to generate summaries and proposals for the user.
Sanitization: The skill enforces a human-in-the-loop requirement, requiring users to explicitly approve or modify any suggested rule changes before they are saved to disk.

rules-distill