agent-orchestration

Pass

Audited by Gen Agent Trust Hub on Sep 21, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill templates facilitate the creation of agents that ingest untrusted user tasks and process them using tools such as web search and database queries. This creates a surface for indirect prompt injection where malicious content in external data sources could influence agent behavior.
  • Ingestion points: Untrusted data enters the agent context via the task parameter in functions such as reactAgent (scripts/agent-workflow-template.ts), multiAgentCollaboration (scripts/agent-workflow-template.ts), and run_support_workflow (scripts/openai-multi-agent.py).
  • Boundary markers: The templates use system prompts to define agent roles and formats, but do not include robust input sanitization or explicit boundary markers to isolate untrusted data.
  • Capability inventory: The agents utilize powerful tools including WebFetch, WebSearch, Read, Glob, Grep, and include template implementations for web searching, database querying, and email sending.
  • Sanitization: No explicit sanitization or filtering of external tool outputs is implemented in the provided template code.
  • [COMMAND_EXECUTION]: The orchestration patterns in references/claude-code-instance-management.md include a coordinator script that utilizes subprocess.Popen to launch and monitor multiple instances of the claude code CLI tool. While intended for parallel scenario execution, this demonstrates the capability to execute system commands.
  • [DYNAMIC_EXECUTION]: The documentation in references/claude-code-instance-management.md instructs users to generate Python scripts (coordinator.py and run_scenario.py) via shell redirects (cat >). This pattern of script generation and subsequent execution is a standard developer workflow for this skill's use-case but involves runtime code creation.
  • [EXTERNAL_DOWNLOADS]: The skill references several external libraries and frameworks, including LangGraph, CrewAI, and the OpenAI Agents SDK. These references target official documentation and repositories of well-known organizations and services.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 21, 2026, 01:02 PM
Security Audit — agent-trust-hub — agent-orchestration