third-party-repo-triage

Pass

Audited by Gen Agent Trust Hub on Nov 3, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to audit untrusted repositories, which creates a significant surface for indirect prompt injection attacks.
  • Ingestion points: All files within external repositories cloned from GitHub (SKILL.md).
  • Boundary markers: The documentation includes a high-level warning to treat third-party repos as "DATOS NO CONFIABLES" (untrusted data), but it does not specify technical delimiters or escaping mechanisms for the ingested content.
  • Capability inventory: The skill possesses capabilities for file system access, network operations (git, curl, gh), and the installation/execution of external code (uv tool install).
  • Sanitization: There is no mention of automated sanitization or filtering of repository content before it is processed by subagents or the main agent.
  • [REMOTE_CODE_EXECUTION]: During the "Phase 2" spike phase, the skill instructs the agent to install external CLI tools using uv tool install and execute them against local data. This allows for the execution of untrusted third-party code on the user's machine.
  • [COMMAND_EXECUTION]: The skill frequently uses shell commands for its core operations, including git clone for repository acquisition, gh api and curl for metadata retrieval, and grep for pattern-based risk scanning.
  • [EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch source code and repository metadata from GitHub, which is a well-known service. These downloads are central to the skill's triaging purpose.
  • [DYNAMIC_EXECUTION]: The skill's reference documentation includes Python code snippets designed to be executed at runtime for parsing package lock files and analyzing dependencies. Additionally, it promotes the dynamic installation and execution of tools discovered during the triage process.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 3, 2026, 10:26 AM