canvas
Pass
Audited by Gen Agent Trust Hub on Feb 28, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill implements an
evalaction that allows the agent to execute arbitrary JavaScript code within the WebView context of connected Mac, iOS, and Android nodes. Furthermore, the 'Live Reload' feature dynamically injects WebSocket client code into static HTML files served by the host. - [COMMAND_EXECUTION]: Documentation provides multiple examples of shell command execution for configuration and network debugging, including the use of
tailscale,jq,lsof, andcryptoclawCLI tools to inspect the environment and manage nodes. - [DATA_EXFILTRATION]: The
snapshotaction enables the capture of screenshots from the canvas view. This capability could be used to visually exfiltrate information rendered on the remote node's display. - [REMOTE_CODE_EXECUTION]: (Indirect Surface) The skill serves HTML/JS content from a local directory (
~/clawd/canvas/) and presents it on remote nodes. If the agent is instructed to render content derived from untrusted external sources, it could lead to indirect injection of malicious scripts into the WebView environment.
Audit Metadata