full-simplify
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-controlled code (diffs) which are analyzed by a chain of agents. Maliciously crafted comments or strings within the analyzed code could influence the agents to perform unintended file modifications.
- Ingestion points: The skill reads the 'subject diff' and 'repository instructions' in Step 1 of the workflow defined in SKILL.md.
- Boundary markers: There are no explicit instructions to use delimiters or security-focused boundary markers to separate the untrusted diff content from the agent prompts.
- Capability inventory: The skill is authorized to edit the working tree (Step 3) and execute shell commands for linting, typechecking, and unit tests (Step 4).
- Sanitization: No sanitization process is defined for the diff content before it is passed to the Moderator and Analyst agents.
- [COMMAND_EXECUTION]: The skill automatically runs linting and unit tests after applying proposals. If an injection attack successfully introduces malicious code into a file or test suite, that code may be executed during this automated check phase.
Audit Metadata