computer-use
Pass
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: References a reference Docker container from Anthropics (ghcr.io/anthropics/anthropic-quickstarts:computer-use-demo-latest) for local testing. This is used for demonstration purposes as part of the official computer use implementation guide.
- [PROMPT_INJECTION]: Explicitly documents the risk of indirect prompt injection where the agent might encounter and follow instructions embedded in screenshots of untrusted web pages or emails. The skill provides specific advice on prompting the agent to ignore such content.
- [COMMAND_EXECUTION]: Mentions the use of a bash tool (bash_20250124) to execute shell commands within the virtualized environment. The skill correctly identifies this as a potential risk factor and advises isolation in disposable sandboxed VMs.
Audit Metadata