worldos-pax-adaptation
Pass
Audited by Gen Agent Trust Hub on Aug 12, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches versioned documents from Google's Firestore API (firestore.googleapis.com). This is a well-known and trusted service, and the associated script (fetch-pax-version.py) includes input validation to prevent identifier injection.
- [COMMAND_EXECUTION]: The agent is instructed to execute local Python scripts (check-pax-adaptation.py, decode-pax-version.py, fetch-pax-version.py) for data validation and transformation. These scripts utilize standard libraries and implement safe parsing for structured JSON data.
- [DATA_EXFILTRATION]: The instructions explicitly forbid the use of private APIs, direct database access, or requesting access tokens in chat, which effectively mitigates common credential harvesting risks.
- [PROMPT_INJECTION]: The skill addresses potential indirect prompt injection from third-party content by requiring the agent to summarize and transform source material rather than reproducing it verbatim. It also uses a mandatory coverage worksheet to audit source decisions. Ingestion points: Firestore snapshots via fetch-pax-version.py; Boundary markers: Explicit instruction to transform source prose; Capability inventory: WorldOS MCP capabilities for simulation management; Sanitization: Script-based decoding of Firestore typed values.
Audit Metadata