The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell scripts (scan-skills.sh and scan-rules.sh) to gather metadata about installed skills and rule files. These scripts use standard Linux utilities (find, grep, awk, jq) and do not perform any destructive operations or unauthorized network access. The use of $HOME and $PWD is constrained to identifying local configuration directories (~/.claude/skills and ~/.claude/rules).
[PROMPT_INJECTION]: While the skill involves a subagent prompt for analysis, the instructions are focused on analytical tasks (matching, extraction, and formatting). There are no instructions to bypass safety filters or override agent behavior in a malicious way. The prompt includes clear criteria and structured output formats to maintain deterministic behavior.
[EXTERNAL_DOWNLOADS]: The skill does not perform any external downloads or fetch remote code. All operations are performed on local files within the agent's expected configuration directories.
[DATA_EXFILTRATION]: No data exfiltration patterns were detected. The scripts read local files for the purpose of rule distillation and present the findings to the user. There are no network operations to send this data to external domains.
[REMOTE_CODE_EXECUTION]: The skill does not involve remote code execution. It uses local scripts for inventory collection, which is a standard pattern for AI agent skills managing local configuration.

rules-distill