cicd-pipeline
Pass
Audited by Gen Agent Trust Hub on May 15, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill instructions demonstrate processing untrusted data, specifically pull request diffs and test logs, by piping them into prompts for automated analysis. This creates a surface for indirect prompt injection where malicious content embedded in the ingested files could influence the agent's automated responses or actions.\n
- Ingestion points: pr-diff.txt and test-results/*.log identified in SKILL.md.\n
- Boundary markers: Absent in the provided examples.\n
- Capability inventory: Examples include capabilities to write pull request comments and create issues via GitHub API.\n
- Sanitization: Not specified for the external data inputs.\n- [COMMAND_EXECUTION]: The TypeScript SDK integration example uses string interpolation within an execSync call. This pattern can lead to command injection if variables like projectDir or prompt are populated from untrusted sources without rigorous sanitization.\n- [EXTERNAL_DOWNLOADS]: The skill includes instructions to download and install the @anthropic-ai/claude-code package from the npm registry. This is an official package from a well-known service provider.
Audit Metadata