impeccable

Fail

Audited by Gen Agent Trust Hub on Nov 21, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONPERSISTENCECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill's primary launcher scripts (scripts/impeccable and scripts/impeccable.cmd) are designed to download executable binaries from a non-vendor GitHub account (github.com/pbakaus/impeccable) and execute them locally. This happens during setup or first run when the platform-specific binary is not already present.
  • [PERSISTENCE]: As documented in reference/hooks.md, the skill installs design detector hooks by modifying agent-specific and IDE configuration files, including .claude/settings.local.json, .cursor/hooks.json, .github/hooks/impeccable.json, and .grok/hooks/impeccable.json. These hooks ensure the external binary is executed automatically whenever the agent edits project files.
  • [COMMAND_EXECUTION]: The skill frequently invokes shell commands via the launcher to run the impeccable binary for auditing, design critique, and "live mode" iteration. The launcher also checks and updates permissions using chmod +x.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user source code for auditing and critique purposes. reference/critique.md and reference/audit.md describe workflows that ingest and analyze project files, representing an indirect prompt injection surface. The skill also performs script injection into local browser sessions during "live mode" (reference/live.md). There is no evidence of boundary markers or sanitization for this ingested content.
  • [PROMPT_INJECTION]: reference/craft.md contains instructions directing the agent not to inform the user about the need to invoke certain commands, which constitutes an attempt to conceal agent behavior or routing logic from the user.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Nov 21, 2026, 10:39 AM