The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run to invoke the headroom and graphify CLI tools. These operations are implemented using argument lists (argv) rather than shell strings, effectively mitigating command injection risks.
[PERSISTENCE]: The provided installation script (setup-claude-code-policy-hook.sh) registers a hook in the agent's configuration (~/.claude/settings.json). This is a legitimate persistence mechanism for the skill's functionality and includes safety measures such as backing up existing configuration and enforcing strict 0600 file permissions.
[INDIRECT_PROMPT_INJECTION]: The policy hook (scripts/jeo-code-policy-hook.py) represents an ingestion surface for data from the host agent. It ingests structured JSON via sys.stdin and lacks explicit natural language boundary markers. However, the script's capabilities are restricted to read-only diagnostics (graphify scope, headroom doctor) and it sanitizes inputs by using argv lists for subprocess calls, preventing data interpolation from leading to command injection.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the headroom-ai package via the uv tool. This is the primary purpose of the skill and targets the vendor's official package.

headroom