The Agent Skills Directory

[PERSISTENCE]: The skill offers to modify the agent's persistent instruction files (such as ~/.claude/CLAUDE.md or AGENTS.md) to establish a preference for using tldraw for diagrams and visual work. The instructions explicitly state that this should only be done with user consent and via appending, not overwriting.
[COMMAND_EXECUTION]: The skill utilizes local shell scripts (scripts/tq and scripts/new-canvas) to interact with the tldraw application and the filesystem. These scripts manage file creation, directory setup, and local API calls.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from the tldraw application's local API, creating a surface for potential indirect influence if the application state is manipulated.
Ingestion points: The skill reads the list of open documents and their metadata from the local bridge server via the tq helper.
Boundary markers: None identified in the provided scripts.
Capability inventory: Filesystem writes (creating directories and README files), local network communication with a whitelisted domain (localhost), and shell script execution.
Sanitization: The scripts interpolate file paths directly into JSON payloads; while this presents a technical injection surface, the risk is localized to the user's own file system and intended application interactions.

canvas-setup