clonedeps

Pass

Audited by Gen Agent Trust Hub on Sep 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content by cloning external repositories into the local workspace for the agent to inspect. This data is then available in the agent's context, creating a surface for indirect prompt injection if the cloned source contains malicious instructions.\n
  • Ingestion points: External code is cloned into .slim/clonedeps/repos/ using git clone (SKILL.md, Step 4).\n
  • Boundary markers: The skill instructions do not explicitly mandate delimiters or 'ignore instructions' warnings when the agent reads the cloned code, though it marks the area as read-only in AGENTS.md.\n
  • Capability inventory: The skill can execute git commands, write to the filesystem (.slim/, .gitignore, AGENTS.md), and perform network operations.\n
  • Sanitization: Includes checks for URL protocols, rejection of embedded credentials, and mandatory user approval before cloning (SKILL.md, Step 3).\n- [COMMAND_EXECUTION]: The workflow involves executing git commands (ls-remote, clone, fetch) to interact with remote repositories. (SKILL.md, Step 4)\n- [EXTERNAL_DOWNLOADS]: The skill is designed to download source code from external URLs (GitHub, GitLab) provided by another tool or the user. (SKILL.md, Step 3 & 4)
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 7, 2026, 11:23 PM
Security Audit — agent-trust-hub — clonedeps