publish-evidence

Pass

Audited by Gen Agent Trust Hub on Aug 26, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it ingests and processes data from external test runs.
  • Ingestion points: Test names, assertion evidence, and test artifacts are ingested from @openwork/testkit outputs into the agent's context for PR reporting.
  • Boundary markers: The skill uses <!-- test-evidence --> as a structural marker in pull request comments, but lacks explicit delimitation or instructions to treat ingested artifact content as untrusted data.
  • Capability inventory: The agent can execute shell commands via pnpm evals:e2e and modify PR comments.
  • Sanitization: No sanitization or validation of the content within the test artifacts is described before it is rendered into the final comment.
  • [COMMAND_EXECUTION]: The skill uses the pnpm package manager to execute local scripts for publishing results.
  • Evidence: The instructions document the use of pnpm evals:e2e --publish with various flags for PR and test run selection.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 26, 2026, 01:21 AM
Security Audit — agent-trust-hub — publish-evidence