implementing-tasks

Pass

Audited by Gen Agent Trust Hub on Mar 10, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill reads and follows instructions ('LLM Prompt') and patterns from plan files in the .plans/ directory, creating a surface for indirect prompt injection.
  • Ingestion points: .plans/ task files, critical-patterns.md, and research agent outputs.
  • Boundary markers: None mentioned.
  • Capability inventory: File editing, code/test generation, test execution, and sub-agent orchestration.
  • Sanitization: No sanitization or validation is applied to instructions extracted from the task files.
  • [COMMAND_EXECUTION]: The skill executes shell-based commands for metadata collection and research activities. Evidence: git diff --stat and research-breadth/depth/technical CLI calls.
  • [COMMAND_EXECUTION]: The skill generates and executes code and test suites based on instructions provided in plan files. Evidence: 'write code + tests, run full suite' instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 10, 2026, 02:24 AM