engineering-workflow-guard
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon data from external mobile repositories, which could potentially contain malicious instructions embedded in the codebase or build logs. \n
- Ingestion points: Repository metadata, file changes, and build/test outputs (SKILL.md). \n
- Boundary markers: Relies on structured procedural handoffs like 'Task contracts' and 'Completion receipts' rather than technical delimiters. \n
- Capability inventory: Shell command execution for builds, Git operations, and file modifications (SKILL.md). \n
- Sanitization: Instructions mandate manual validation of user input and authorization on trust boundaries. \n- [SAFE]: The instructions contain strong negative constraints against exposing sensitive information such as API keys, signing certificates, or private user data in any artifacts, logs, or prompts. \n- [SAFE]: The skill enforces a policy of explicit authorization for destructive actions (e.g., force-push, branch deletion) and external publishing, reducing the risk of unauthorized autonomous changes.
Audit Metadata