The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from potentially untrusted sources such as the local codebase, GitHub PR review comments, and Claude conversation history. \n
Ingestion points: Reads codebase files (Step 3), fetches PR comments via GitHub API (Step P3), and parses session history from .jsonl files (Step C3). \n
Capability inventory: The skill has 'Write' access to the local filesystem and uses various 'Bash' tools for analysis. \n
Sanitization: It includes a dedicated 'Security Self-Check' (Step 6.5) that greps for potential secrets, high-entropy strings, and internal URLs in the generated rule files, redacting any findings before completion. \n- [DATA_EXPOSURE]: The skill accesses sensitive local configuration and history files, such as ~/.claude/settings.json and session logs in ~/.claude/projects/. \n
This access is required for its core functionality of extracting rules from previous conversations and resolving project settings. \n
The data is processed locally to generate documentation and is not exfiltrated to external domains. \n- [COMMAND_EXECUTION]: The skill uses node to execute its own bundled JavaScript utility, extract_session_messages.mjs. \n
This script is a specialized parser for Claude session logs and does not perform network operations or execute arbitrary external code.

extract-rules