bulk-generate-concepts

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [Indirect Prompt Injection] (HIGH): The skill processes untrusted data from external websites (ministry sites) and saves it to a research document that dictates the scope and content for sub-agents and PR descriptions.
  • Ingestion points: Step 3 involves researching external curriculum standards from the web and extracting learning objectives and terminology.
  • Boundary markers: Absent. There are no instructions to delimit the researched content or warn sub-agents to ignore embedded instructions within the research doc.
  • Capability inventory: The skill possesses significant capabilities including file system writes, shell command execution (git, bun), and external platform interaction (gh pr create).
  • Sanitization: None. The researched content is used directly to create branch names, commit messages, and PR bodies.
  • [Command Execution] (MEDIUM): The skill executes multiple shell commands (git, gh, bun run). While these are standard for a development workflow, they are parameterized with variables derived from external research data (e.g., {subject}, {N}, {curriculum}), which could be manipulated via indirect injection.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 12:47 PM