agentspace-human-agent-workspace
Pass
Audited by Gen Agent Trust Hub on Jul 2, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides instructions to download the AgentSpace codebase from a public GitHub repository (https://github.com/HKUDS/AgentSpace.git) and mentions a hosted platform at hire-an-agent.online.- [COMMAND_EXECUTION]: The installation guide requires executing shell commands for setup and development, including
npm install,npm run setup, andnpm run dev:web.- [PROMPT_INJECTION]: The multi-agent orchestration architecture creates a surface for indirect prompt injection by processing external data through automated agents. - Ingestion points: Agents ingest data from shared communication channels, workspace document repositories, and integrated Google Workspace files.
- Boundary markers: The provided code examples and instructions do not define explicit boundary markers or delimiters to separate untrusted data from agent instructions.
- Capability inventory: Agents possess broad capabilities including document read/write, multi-agent task delegation, and execution of runtime-specific skills.
- Sanitization: The orchestration logic demonstrated in the skill does not include evidence of input sanitization or validation before data is passed to agent runtimes.
Audit Metadata