arbor
Pass
Audited by Gen Agent Trust Hub on Aug 17, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to iteratively optimize external artifacts such as codebases and data pipelines. This creates a surface for indirect prompt injection where malicious instructions embedded in the processed artifacts could attempt to influence the agent's behavior or subagent execution.
- Ingestion points: External artifact repositories, data files, and evaluator outputs.
- Boundary markers: The skill utilizes a structured executor brief template to constrain subagent tasks.
- Capability inventory: Subagents are granted access to Bash and Agent tools to run evaluators and implement changes.
- Sanitization: Absent; the skill does not perform sanitization on artifact content as its primary purpose is code modification and execution.
- [EXTERNAL_DOWNLOADS]: The skill provides instructions for installing the 'uv' package manager and cloning the official Arbor research repository. Evidence: References to docs.astral.sh and github.com/RUC-NLPIR/Arbor. These are well-known technology and academic sources used for environment setup rather than automated malicious downloads.
Audit Metadata