browser-runtime
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes patterns for ingesting and processing untrusted data from external sources (Web APIs and browser storage).
- Ingestion points: Untrusted data enters the agent context through fetch responses and localStorage/sessionStorage reads as described in SKILL.md.
- Boundary markers: The instructions explicitly state to 'validate untrusted decoded payloads before trusting their TypeScript type' and to use 'explicit ownership rules' for state.
- Capability inventory: No scripts or executable code provided; the skill purely defines instructional constraints. It does not perform subprocess calls, arbitrary file writes, or network operations beyond the standard browser context defined in its scope.
- Sanitization: The skill recommends validation of payloads, explicit serialization, and versioning of persisted shapes to prevent issues with malformed or malicious data.
- [NO_CODE]: The skill consists entirely of documentation and evaluation configuration without any executable scripts or binary files.
Audit Metadata