agentspace-human-agent-workspace
Warn
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill requires cloning a source code repository from
https://github.com/HKUDS/AgentSpace.gitand installing dependencies vianpm installduring the setup process. - [COMMAND_EXECUTION]: The instructions involve executing several shell commands, including
npm run setup,npm run dev:web,npm run db:migrate, and a suite ofagentspaceCLI commands for managing workspace resources and permissions. - [INDIRECT_PROMPT_INJECTION]:
- Ingestion points: The skill is designed to ingest and process untrusted data from collaborative channels, document knowledge bases, and Google Workspace integrations.
- Boundary markers: There are no defined boundary markers or explicit instructions provided to the agent to ignore or isolate instructions that may be embedded within processed documents or chat messages.
- Capability inventory: The skill manages agents capable of executing tasks across various runtimes and performing high-privilege operations such as
kubectl-apply(via the permission manager). - Sanitization: No evidence of sanitization or filtering of external content is presented in the provided logic.
Audit Metadata