sequence-diagram
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local shell scripts, specifically
scripts/new-canvasandscripts/tq, to automate the creation, opening, and programmatic update of .tldraw files through the local tldraw server. - [DATA_EXFILTRATION]: The skill reads application-specific configuration files (server.json) from standard local paths (e.g.,
~/Library/Application Support/tldraw/) to retrieve session tokens required for authenticating with the local tldraw API. - [DYNAMIC_EXECUTION]: The skill constructs JavaScript code blocks at runtime and sends them to the tldraw application's execution endpoint to perform automated canvas updates and layout adjustments.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input, such as prose or DSL specifications, to build diagram models, which represents a potential surface for prompt injection attacks.
- Ingestion points: User-provided diagram descriptions or DSL specs processed in SKILL.md.
- Boundary markers: No specific delimiters or safety warnings for embedded instructions are identified.
- Capability inventory: Execution of local shell scripts, network requests to the local tldraw server, and application-level JavaScript execution.
- Sanitization: The skill relies on structural parsing but does not include explicit sanitization or filtering of the input data.
Audit Metadata