write-review-apply
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [DYNAMIC_EXECUTION]: Step 2 of the workflow instructs the agent to execute commands, code changes, or regex patterns supplied by an article review in a temporary directory to verify their validity. This involves running untrusted code snippets derived from potentially external data.
- [INDIRECT_PROMPT_INJECTION]: The skill processes review reports which are considered untrusted data. A malicious review could contain instructions to exploit agent capabilities. 1. Ingestion points: Review reports from conversation or saved files in Step 1. 2. Boundary markers: No specific delimiters defined for review content. 3. Capability inventory: Shell command execution, file writing, and code verification. 4. Sanitization: Verification and change-checking steps do not filter for adversarial instructions.
- [COMMAND_EXECUTION]: The skill executes several external shell scripts and Python utilities located in peer directories using relative paths to manage article status and validation.
Audit Metadata