plan-to-task-list-with-dag

Pass

Audited by Gen Agent Trust Hub on Mar 15, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill instructions do not contain any malicious prompt injection patterns, obfuscation, or hardcoded credentials. The workflow is focused on codebase exploration and task planning using standard MCP tools.
  • [INDIRECT_PROMPT_INJECTION]: The skill has an inherent attack surface for indirect prompt injection as it ingests untrusted data from the codebase during its 'Explore' phase (using mcp__codemap tools) and interpolates this data into task descriptions.
  • Ingestion points: Codebase content retrieved via mcp__codemap__search_code and mcp__codemap__get_file_summary in SKILL.md.
  • Boundary markers: None identified; codebase content is directly used to build mental models and task descriptions.
  • Capability inventory: The skill can write files to the local file system (plans/ directory) and interact with the user via AskUserQuestion.
  • Sanitization: No explicit sanitization or escaping of codebase content is mentioned before it is included in the generated markdown or JSON plans.
  • [EXTERNAL_DOWNLOADS]: While documentation examples mention tasks that might involve network operations (e.g., Linear API integration via fetch), the skill itself does not perform any external network requests or downloads.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 15, 2026, 01:29 PM