arbor

Pass

Audited by Gen Agent Trust Hub on Jul 31, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The documentation in references/arbor-upstream.md provides instructions for cloning and installing the Arbor CLI from a third-party GitHub repository (github.com/RUC-NLPIR/Arbor).\n- [COMMAND_EXECUTION]: The skill executes user-provided shell commands for development and test evaluation (E_dev and E_test) to score artifacts during the optimization cycles.\n- [PROMPT_INJECTION]: The skill architecture is susceptible to indirect prompt injection.\n
  • Ingestion points: The coordinator agent ingests insight and result data from the hypothesis tree (managed by scripts/tree.py), which is populated by feedback from subagents.\n
  • Boundary markers: A structured return format is defined in references/executor-brief.md, though the insight field accepts free-form text.\n
  • Capability inventory: The skill possesses the ability to execute shell commands (Bash), orchestrate subagents (Agent), and modify files (Edit) across the environment.\n
  • Sanitization: The skill does not perform sanitization of natural language insights before they are processed by the coordinator agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 31, 2026, 08:20 PM
Security Audit — agent-trust-hub — arbor