pr-brief
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill frequently executes
gitandgh(GitHub CLI) commands to retrieve repository state and modify Pull Requests. It usesgit logandgit diffto gather context andgh pr editto apply the generated body and labels. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the repository's history and PR metadata, which could be used to manipulate the agent's output.
- Ingestion points: Data is ingested from
git diff,git log(including commit subjects and bodies), existing PR descriptions, and various markdown files within the repository (e.g.,CODING_STANDARDS.md,CONTRIBUTING.md). These are identified inSKILL.mdStep 2 andreferences/assessor-brief.md. - Boundary markers: The skill employs a subagent for assessment (
references/assessor-brief.md) with instructions to remain read-only and a specific directive to validate the safety of the change ('your job is to check'). However, the instructions do not include rigorous escaping or delimiters for the untrusted text being processed. - Capability inventory: The skill is capable of modifying PR descriptions and labels via
gh pr edit. It does not have the capability to push code, merge branches, or interact with external network services outside of the GitHub API. - Sanitization: There is no evidence of sanitization or filtering for the content harvested from git logs or diffs before it is interpolated into the prompts for the assessor or the final PR body.
Audit Metadata