triaging-visual-review-runs

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests and processes untrusted external data from GitHub pull requests, storybook/playwright test configurations, and snapshot elements. Ingestion points: Data enters the context via PR fields, branch names, commit SHAs, and screenshot URLs. Boundary markers: Absent; there are no explicit delimiters or structural controls separating untrusted input data from the instructions. Capability inventory: The skill has access to state-altering capabilities including posthog:visual-review-runs-finalize-create, which can alter branch state and green GitHub status gates. Sanitization: Absent; no validation or input filtering mechanisms are implemented.
  • [COMMAND_EXECUTION]: The visual inspection workflow explicitly prompts the execution of shell commands using curl to pull snapshot artifacts to the local /tmp directory.
  • [EXTERNAL_DOWNLOADS]: The skill directs the agent to download remote image assets via dynamic pre-signed artifact URLs provided during runtime execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 03:19 PM
Security Audit — agent-trust-hub — triaging-visual-review-runs