Generating New Skills

Pass

Audited by Gen Agent Trust Hub on Aug 1, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill establishes a process for the agent to ingest user-provided methodology and codify it into new, persistent instructions. This presents an indirect prompt injection risk.
  • Ingestion points: The skill monitors user requests, workflow explanations (2+ occurrences), and explicit commands like 'remember this'.
  • Boundary markers: The skill provides no instructions or delimiters to distinguish between safe workflows and potentially malicious instructions within the learned content.
  • Capability inventory: The skill allows the agent to write new markdown files to the .claude/skills/ directory, which are then treated as authoritative instructions for future tasks.
  • Sanitization: There is no evidence of logic to sanitize, validate, or filter user input before it is transformed into a stored skill.
  • [COMMAND_EXECUTION]: The skill documentation explicitly instructs the agent to use shell commands to manage its state.
  • Evidence: The skill includes code blocks suggesting the execution of npm run learn:track and npm run learn:record to persist learned patterns.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 1, 2026, 10:46 AM
Security Audit — agent-trust-hub — Generating New Skills