om-ai-context-maintenance
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a bundled Python script (
scripts/rotate-runs.py) to perform local archival. The script is restricted to the Python standard library and includes robust path validation to prevent directory traversal or symlink attacks. - [PROMPT_INJECTION]: The skill explicitly defines an 'Untrusted content boundary' in both
SKILL.mdandreferences/agentic-setup.md. It instructs the agent to treat all run artifacts, names, and tracker text as data and to ignore any instructions found within them (e.g., 'ignore previous instructions'). The static detector's flag onagentic-setup.mdis a false positive as the file correctly documents how to handle injections rather than performing one. - [DATA_EXFILTRATION]: The skill has no network access beyond read-only tracker operations (GitHub/GitLab PR lists). It explicitly prohibits printing secrets or artifact contents in reports.
- [DYNAMIC_EXECUTION]: The skill uses
shutil.rmtreeandos.unlinkonly after a multi-step verification process (SHA-256 integrity checks of the created archive) and only within an isolated PR worktree, never on the user's active checkout.
Audit Metadata