activity-diagram
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill reads the
server.jsonfile from the application's local data directory to retrieve the port and authentication token for local server communication. All network operations are directed tolocalhost.\n- [COMMAND_EXECUTION]: The skill executes shell scripts (scripts/tqandscripts/new-canvas) to copy templates, open the tldraw app, and send commands viacurl.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied workflow descriptions. It mitigates injection risks by parsing the input into a structured model before generating canvas commands.\n - Ingestion points: User-provided prose or lists in the prompt (SKILL.md).\n
- Boundary markers: The workflow requires parsing into a normalized model (
{ lanes, nodes, edges }) before execution.\n - Capability inventory: JavaScript execution via
tq, file creation for new diagrams, and local network requests.\n - Sanitization: The use of a structural intermediate model prevents raw user input from being executed directly in the tldraw environment.
Audit Metadata