vibe-delegate

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The relay.mjs script manages the implementation loop by executing the vibe CLI and git. It also uses taskkill on Windows and process.kill on Unix for process management. A secondary script lane.mjs may be executed if the --lane flag is used, referencing a local file within the same vendor's skill set.
  • [EXTERNAL_DOWNLOADS]: The skill documentation provides instructions for installing uv (Astral) and mistral-vibe (Mistral AI) via official channels. These are recognized tools from established organizations.
  • [INDIRECT_PROMPT_INJECTION]: The skill acts as a proxy, passing user-generated task briefs to a downstream agent (vibe).
  • Ingestion points: Briefs are ingested through the --brief file argument or via stdin in relay.mjs.
  • Boundary markers: The writing-the-brief.md reference recommends using XML-style tags to delimit task components.
  • Capability inventory: The relay script can trigger file modifications and arbitrary command execution (when using --full-access) via the vibe CLI.
  • Sanitization: No content-based sanitization is performed on the brief, though length limits are enforced for platform compatibility.
  • [DATA_EXPOSURE]: The documentation explicitly warns that because the brief is passed to the CLI via a command-line argument, it remains visible in the host's process list. This is standard behavior for CLI tools, and the skill provides guidance to keep secrets out of the brief text.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 06:29 PM
Security Audit — agent-trust-hub — vibe-delegate