The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes Python scripts (profile_claude_md.py, sink_sections.py, markdown_headings.py) designed to be executed by the agent to analyze and modify files. The sink_sections.py script performs file system operations, including reading source content, appending it to target reference files, and creating timestamped backups. A test suite in tests/test_scripts.py uses subprocess.run to execute these local scripts for regression testing, which is a standard and safe development practice.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process user-controlled documentation files (CLAUDE.md, AGENTS.md). These files serve as ingestion points for untrusted data. The skill mitigates this risk by employing deterministic Python scripts to parse and move sections based on heading structures, rather than relying on the LLM's interpretation of the content during the move, which reduces the possibility of an attacker using document content to hijack the optimization process.
[DYNAMIC_EXECUTION]: In references/verification-recipes.md, the skill provides shell command templates (recipes) that extract and execute hook commands from the user's local ~/.claude/settings.json file. This is intended as a diagnostic tool for the user to verify their existing environment configurations and does not facilitate the execution of unknown or remote code.
[SAFE]: The sink_sections.py script implements a robust security check by verifying that target paths do not resolve to symlinks outside the intended directory. This prevents the agent from accidentally writing to sensitive system files or other repositories that may be symlinked into the project structure.

claude-md-progressive-disclosurer