tasklist
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data when importing action items from project artifacts such as meeting notes, site reports, or plan documents. This creates a vulnerability surface where instructions embedded in those files could attempt to influence the agent's task creation behavior.
- Ingestion points: The import operation reads content from user-specified project files (e.g., meetings/, site-reports/, or plans/).
- Boundary markers: The skill implements mitigation by requiring mandatory human-in-the-loop validation, explicitly stating 'Preview candidates before mutation' and 'Ask the user to select individual candidates.'
- Capability inventory: The agent is restricted to standard file operations (Read, Write, Edit, Glob, Grep) and must use AskUserQuestion for confirmations.
- Sanitization: The skill enforces a strict Markdown table schema for TASKS.md and normalizes all source paths and item IDs to ensure traceability and prevent silent duplication.
Audit Metadata