consult-uat-writer
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests external PRD documents from Google Drive, which represents an indirect prompt injection surface. However, this is significantly mitigated by the skill's design:
- Ingestion points: The skill reads PRD files using Google Workspace integration tools (Step 1).
- Boundary markers: The skill uses strict format parsing, searching for four specific Markdown structures (H1 headers, Size/DoD lines, AC pairs, and Out-of-scope sections). Deviation from this format causes the skill to stop (Hard Rule 1).
- Capability inventory: The skill possesses the ability to perform SQL writes to a CRM database (Supabase), create and move Google Docs, and create Gmail drafts.
- Sanitization: A mandatory manual approval gate (Step 4) requires the user to review the generated scenario table before any persistent changes are made to the CRM or filesystem. This ensures human oversight over the ingested data's transformation.
- [COMMAND_EXECUTION]: The skill performs SQL operations against a Supabase database to log tasks and update deal notes. These operations utilize predefined SQL templates and are routed through an MCP tool, representing standard interaction with the vendor's CRM infrastructure.
Audit Metadata