write-pr-description
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local commands using
gitand the GitHub CLI (gh) to analyze branch diffs, check authentication status, push branches, and create pull requests. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources like git diffs, commit messages, and repository templates, which could contain instructions intended to influence the agent's behavior.
- Ingestion points: Branch diffs, commit subjects, branch names, and repository files such as
.github/PULL_REQUEST_TEMPLATE.md(SKILL.md). - Boundary markers: None explicitly mentioned for interpolation, though the skill instructs the agent to verify that generated tokens appear verbatim in the source diff.
- Capability inventory: The skill can perform network-facing actions including
git pushandgh pr create(SKILL.md). - Sanitization: The skill implements a verification step (Step 5) to ensure claims in the PR body match the actual diff and requires user confirmation before execution (Step 6).
Audit Metadata