reverse-tornado-okr
Pass
Audited by Gen Agent Trust Hub on Jul 25, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes bundled shell and Python scripts (e.g., okra-store.sh, check-plain-wording.py) to manage the local OKR state, perform environment pre-flight checks, and verify artifact completeness against provided schemas. These operations are strictly local to the workspace.
- [EXTERNAL_DOWNLOADS]: References a PDF hosted by a well-known service (OpenAI) for documentation and provenance. The instructions specifically command the agent not to fetch this URL at runtime, instead relying on a provided text extract.
- [DATA_EXFILTRATION]: State, metrics, and logs are maintained in a local .okra directory. The skill contains clear directives to keep all repository content, evidence, and credentials within the local environment unless the user explicitly requests an external model path.
- [PROMPT_INJECTION]: The skill implements a defensive 'anti-goal' system and multi-agent role-splitting (Orchestrator/Worker) designed to prevent model hallucinations and ensure that goals are satisfied through objective evidence rather than model self-certification.
Audit Metadata