ship-it
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill autonomously interacts with 'live surfaces' (such as URLs, bot interfaces, and APIs) during its Step 5 verification loop, which introduces a surface for instructions embedded in external content to influence agent behavior.
- Ingestion points: The skill ingests data from user-supplied feature descriptions, internal plan files (
docs/qa-checklist.yaml), and external data fetched from the web or APIs during live testing. - Boundary markers: Absent. There are no instructions defining boundaries or 'ignore' rules to prevent the agent from following malicious instructions found within the code or external data it processes during the live-QA loop.
- Capability inventory: The skill possesses extensive capabilities including file modification (Write/Edit), system command execution (Bash), and automated web interaction tools like Playwright or
/browse. - Sanitization: Absent. The skill does not implement validation, escaping, or cleaning of external data before it is used to drive autonomous agent actions.
- [COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute multiple shell commands to manage git branches (e.g.,
git checkout -b), run build processes, and perform automated testing autonomously without requiring user confirmation for individual steps.
Audit Metadata