add-rules

Pass

Audited by Gen Agent Trust Hub on Oct 27, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external 'policy sources' (local files or remote URLs) which are compiled into rules. This creates an attack surface where malicious instructions embedded in the source documents could influence the agent or the target system.
  • Ingestion points: The instruction parameter and the sources array (containing local paths or remote URLs) defined in the SKILL.md workflow.
  • Boundary markers: The instructions advise the agent to 'Preserve the source's obligations... without inventing policy' and to 'Compile durable constraints only', but no technical sanitization is specified.
  • Capability inventory: The skill uses client.rule_compile (SDK) or POST /api/v1/rules/compile (HTTP) to send data to the OpenViking service, and client.read_raw or GET /api/v1/content/read to retrieve results.
  • Sanitization: There is no explicit mention of sanitizing or escaping the content fetched from external sources before processing.
  • [DATA_EXFILTRATION]: The skill workflow involves reading local files (e.g., /local/path/policy.md) and uploading their content to an external API (OpenViking) via POST /api/v1/resources/temp_upload or SDK methods. While intended for policy compilation, this mechanism could be misused to exfiltrate sensitive local data if the agent is directed to unauthorized paths.
  • [EXTERNAL_DOWNLOADS]: The skill is designed to fetch remote content from user-provided URLs in the sources field (e.g., https://example.org/policy.md) to be used as input for the compilation process.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 27, 2026, 09:09 AM