The Agent Skills Directory

[SAFE]: The skill implements a mandatory audit phase before any cleanup, ensuring the user is presented with a clear inventory of potential targets, estimated savings, and associated risks.\n- [SAFE]: The bundled audit script (scripts/audit_codex_home.py) uses Python's standard library to perform read-only file and process analysis. It employs secure practices, such as copying SQLite databases to temporary directories before running integrity checks to avoid interfering with active processes.\n- [SAFE]: Process discovery is intentionally limited to executable names, omitting command-line arguments and environment variables to prevent the accidental exposure of sensitive tokens or paths.\n- [SAFE]: The skill provides explicit instructions to preserve sensitive data, specifically forbidding the deletion of credentials, configuration files, and plugin metadata.\n- [SAFE]: Maintenance workflows, such as log database compaction, include mandatory integrity checks, logical content baselines, and recovery backups, mitigating the risk of data loss during physical maintenance tasks.\n- [SAFE]: The skill is configured for explicit invocation only, preventing the agent from autonomously triggering cleanup actions during unrelated tasks.

codex-cleanup