weigh-comments
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from pull request review threads and comments fetched via the GitHub API (SKILL.md, Step 1). External contributors could embed malicious instructions within these comments to influence the agent's analysis or actions.
- [INDIRECT_PROMPT_INJECTION]: Mandatory Evidence Chain:
- Ingestion points: PR comments and review nodes fetched in SKILL.md via
gh api graphql. - Boundary markers: Absent; there are no instructions to delimit or ignore instructions embedded within the fetched text.
- Capability inventory: The skill possesses the ability to read/write files, push to git branches, file tickets, and execute commands as part of the "Verdict" phase.
- Sanitization: Absent; the skill does not specify any sanitization or validation of the comment body before the agent processes or acts upon it.
- [DYNAMIC_EXECUTION]: In Step 2, the instructions direct the agent to "run the case it [the comment] describes when you can." This creates a risk where the agent might interpret and execute dangerous shell commands or malicious code snippets if they are presented as a "test case" or "reproduction step" in a review comment.
- [COMMAND_EXECUTION]: Uses the
gh(GitHub) CLI tool to perform repository queries and metadata retrieval, which is standard behavior for this type of tool but involves shell execution context.
Audit Metadata