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
gitandjjcommands. - 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