github-evidence-kit

Pass

Audited by Gen Agent Trust Hub on Jun 7, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests and processes forensic data from external APIs and JSON stores which are then presented in the agent's context.
  • Ingestion points: Forensic evidence is loaded from JSON files via EvidenceStore.load (src/store.py) or collected from external APIs (GitHub, GH Archive, Wayback Machine) via modules in src/collectors/.
  • Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are used when interpolating external content into the agent's prompt context.
  • Capability inventory: The skill possesses the capability to perform network requests via requests.get (src/verifiers/consistency.py), execute system commands via subprocess.run (src/clients/git.py), and write to the local file system via EvidenceStore.save (src/store.py).
  • Sanitization: The ConsistencyVerifier in src/verifiers/consistency.py performs HTTP GET requests on verification URLs provided in evidence objects without pre-validating them against a whitelist, creating a potential SSRF surface.
  • [COMMAND_EXECUTION]: The Git client wrapper in src/clients/git.py interacts with local repositories using subprocess.run. It employs security-focused command-line arguments (_SAFE_GIT_OVERRIDES) such as -c core.fsmonitor= and -c core.editor=true to mitigate exploitation risks from malicious repository configurations.
  • [EXTERNAL_DOWNLOADS]: The skill fetches data from established technology services including the GitHub API, Google Cloud BigQuery, and the Wayback Machine. All dependencies listed in requirements.txt are well-known libraries.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 7, 2026, 01:29 PM