symmetric-dogfooding

Pass

Audited by Gen Agent Trust Hub on Oct 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill defines workflows where an AI agent processes outputs from external repositories to perform validation. This creates an attack surface where untrusted content from a repository could attempt to influence the agent's behavior during the validation process.
  • Ingestion points: Integration tests in SKILL.md (Implementation Guide Phase 3) and example-setup.md ingest "real outputs" from partner repositories.
  • Boundary markers: The provided templates do not define specific content delimiters or instructions to isolate external data from agent instructions.
  • Capability inventory: The skill uses Read, Write, and Bash tools to execute test suites and manage repository files.
  • Sanitization: No specific sanitization or schema validation of external content is mandated before it is processed by the testing framework.
  • [EXTERNAL_DOWNLOADS]: The skill templates and implementation examples reference external Git repositories for dependency management during the testing phase.
  • Evidence: References to repositories under the terrylica organization in example-setup.md (e.g., https://github.com/terrylica/rangebar-py) represent legitimate vendor-owned resources used for integration testing examples.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 8, 2026, 02:19 AM