The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides an interface to execute arbitrary JavaScript code on the local tldraw application server via the /exec and /api/search endpoints.
Evidence: The scripts/tldraw_api.sh script uses curl to POST JavaScript code from local files or strings to http://127.0.0.1:[PORT]/api/doc/[ID]/exec.
[REMOTE_CODE_EXECUTION]: The skill implements dynamic loading to hook into the application's internal renderer modules at runtime.
Evidence: The desktop_bridge_prefix in scripts/tldraw_api.sh uses await import(src) where the path is discovered dynamically by searching the DOM for script tags matching specific patterns.
[PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by ingesting and processing content from tldraw documents that may originate from untrusted sources.
Ingestion points: scripts/tldraw_api.sh (via shapes, bindings, and serialize commands) and references/api.md (via api.getShapes).
Boundary markers: No delimiters or explicit warnings are provided when the agent reads document data.
Capability inventory: The skill can execute arbitrary JS in the application, write/save files to the local system, and capture screenshots.
Sanitization: While the script safely quotes arguments using jq when building JavaScript strings, it does not sanitize the content retrieved from documents before presenting it to the agent's context.

tldraw-api