using-loopkit

Pass

Audited by Gen Agent Trust Hub on Jul 15, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill employs strong imperative language and explicit override commands to direct agent behavior.
  • Evidence: "EXTREMELY-IMPORTANT", "YOU MUST USE IT", and "This overrides 'just answer quickly' instincts. Not negotiable."
  • While intended to enforce a development workflow, these patterns are similar to techniques used to bypass standard agent response characteristics.
  • [INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface by instructing the agent to load and execute instructions from a directory of external files.
  • Ingestion points: The agent is directed to read .claude/skills/<name>/SKILL.md based on trigger phrases found in the content it processes.
  • Boundary markers: The skill does not provide delimiters or warnings to treat the content of these external files as untrusted.
  • Capability inventory: The agent is instructed to "follow the skill exactly," granting the logic within those external files significant influence over agent actions.
  • Sanitization: No validation or sanitization is performed on the content of the loaded skill files.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 15, 2026, 11:55 PM
Security Audit — agent-trust-hub — using-loopkit