The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/audit_codex_home.py script executes the system ps command using subprocess.run to build a process tree of running applications. This command uses a static set of flags (-axo pid=,ppid=,rss=,comm=) and does not interpolate user-supplied input, ensuring it is limited to its intended discovery function.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a potential attack surface by ingesting local filesystem metadata that could contain adversarial instructions. Ingestion points: Directory listings and file metadata read by the audit_codex_home.py script. Boundary markers: Explicitly defined in the 'Preserve these boundaries' section of SKILL.md, which mandates read-only planning. Capability inventory: The skill allows for file deletion and process termination based on audit results. Sanitization: While no programmatic string sanitization is present, the skill requires mandatory human confirmation and validation against established storage contracts before any mutation occurs, mitigating risk.
[EXTERNAL_DOWNLOADS]: The skill documentation (references/storage-contract.md) includes links to official OpenAI documentation and GitHub repositories. These references are used for informational purposes regarding application storage semantics and are hosted by a trusted organization.

codex-cleanup