prove

Fail

Audited by Gen Agent Trust Hub on Sep 18, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill provides instructions to download and execute the Lean version manager (elan) installation script directly from the official LeanProver GitHub repository via a shell pipe.
  • [EXTERNAL_DOWNLOADS]: The skill initiates a download of the Lean Mathematical Library (Mathlib), approximately 2GB in size, from official repositories during the initial build phase.
  • [COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute environment-specific commands, including lake for project management and proof verification, and loogle-search for querying theorem signatures.
  • [INDIRECT_PROMPT_INJECTION]:
  • Ingestion points: The skill retrieves data from external web sources and specialized search engines (Nia, Perplexity) during the Phase 1 Research stage (SKILL.md).
  • Boundary markers: External search results are incorporated into the agent's context without specific delimiters to isolate potentially malicious instructions.
  • Capability inventory: The agent possesses Bash, Write, and Edit capabilities, allowing it to generate and execute Lean code based on the information gathered from external sources.
  • Sanitization: No explicit validation or filtering of retrieved web content is performed before it is used to draft theorem statements or proof skeletons.
Recommendations
  • HIGH: Downloads and executes remote code from: https://raw.githubusercontent.com/leanprover/elan/master/elan-init.sh - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Sep 18, 2026, 05:11 PM
Security Audit — agent-trust-hub — prove