internet-court

Warn

Audited by Gen Agent Trust Hub on Jul 22, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructions explicitly direct the agent to fetch sub-skills and their corresponding logic from a remote repository.
  • Evidence: Found in SKILL.md: "...let this router fetch them on demand from the repo" and "...load the specific sub-skill's SKILL.md from the package (on disk, or from the repo's raw URL)".
  • [REMOTE_CODE_EXECUTION]: Loading remote instructions (SKILL.md files) from an unverified source allows external content to dictate agent logic.
  • Evidence: The agent is instructed to "never invent a protocol's behavior" and instead rely on the mechanics of dynamically fetched SKILL.md files from an external repository.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface by ingesting instructions from external sources without defined sanitization or boundary markers.
  • Ingestion points: Remote fetching of SKILL.md files from https://github.com/internet-court/internet-court-skill.
  • Boundary markers: Absent; the skill does not specify delimiters or warnings to ignore embedded instructions in the fetched files.
  • Capability inventory: The skill is designed to handle payment processing (x402), escrow, delegated permissions (ERC-7710/7715), and dispute resolution.
  • Sanitization: Absent; no mention of validating or escaping the content retrieved from the remote URL before it is processed by the agent.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 22, 2026, 06:59 PM
Security Audit — agent-trust-hub — internet-court