cherrypicking-pr-to-v3

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill directs the agent to use command-line version control tools, specifically Git and Jujutsu (jj), to duplicate commits and manage branch state.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external source code from Pull Requests and commits, which serves as a potential vector for malicious instructions embedded in code comments or strings to influence the agent's translation behavior.
  • Ingestion points: Reads commit content and source code via git and jj commands.
  • Boundary markers: None provided in the instructions.
  • Capability inventory: File system writes, branch manipulation, and command execution via Git/Jujutsu.
  • Sanitization: None identified; relies on the agent's internal logic during the translation checklist process.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 04:08 AM
Security Audit — agent-trust-hub — cherrypicking-pr-to-v3