impeccable

Pass

Audited by Gen Agent Trust Hub on Apr 13, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as several of its components (such as 'teach-impeccable', 'normalize', and 'extract') are designed to read and interpret content from the user's local repository to identify design patterns, tokens, and existing components. If a file in the project contains malicious instructions disguised as code or documentation, the agent might inadvertently follow those instructions during the design audit. \n
  • Ingestion points: codebase scanning in 'teach-impeccable/SKILL.md' (Step 1), 'normalize/SKILL.md' (Plan Step 1), and 'extract/SKILL.md' (Discover Step 1). \n
  • Boundary markers: The skill does not provide instructions to wrap ingested code in delimiters or to ignore embedded natural language instructions. \n
  • Capability inventory: The skill performs file reads and writes a synthesized design context to 'AGENTS.md' in the project root. \n
  • Sanitization: No sanitization or validation of the ingested codebase content is specified.\n- [COMMAND_EXECUTION]: The skill instructs the agent to use shell tools like 'grep' to search the codebase for design system documentation and component implementations. While this is a standard and necessary behavior for the skill's purpose, it involves direct interaction with the host file system based on potentially unverified patterns.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 13, 2026, 03:36 PM