ship-pr
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content by reading code diffs and untracked files during the review process (SKILL.md Step 3 and references/review-prompt.md Step 2). This exposes the agent to potential instructions embedded in source code changes. \n
- Ingestion points: Code diffs (git diff) and untracked files (git status) processed in the fresh-context review step. \n
- Boundary markers: Present. The instructions mandate spawning a reviewer with no memory of the implementation conversation and require the reviewer to read files directly rather than accepting provided diffs or summaries. \n
- Capability inventory: The skill executes local shell commands via
pnpm,git, andgh. \n - Sanitization: None explicitly mentioned for the source code content being analyzed. \n- [COMMAND_EXECUTION]: The skill uses
pnpmfor linting and testing,gitfor branch management and diffing, andghfor pull request creation (SKILL.md Steps 1 and 4). These are standard operations for development workflows within the tester-army/e2e repository environment.
Audit Metadata