The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct the agent to install the @kitschpatrol/tldraw-cli package globally using npm install -g. It also includes steps to download specific versions of the Chrome browser binary using npx puppeteer browsers install chrome@<version> to support headless rendering via Puppeteer.
[COMMAND_EXECUTION]: The workflow involves multiple shell command executions to verify tool availability, create directories, and perform the export process. This includes commands such as tldraw --version, mkdir -p, tldraw export, and operating system-specific commands like open, xdg-open, or start to display the resulting files to the user.
[INDIRECT_PROMPT_INJECTION]: The skill implements a vision-based self-check mechanism (Step 5) where an agent-generated image, based on user-provided natural language descriptions, is processed by a vision-enabled model. This creates a potential surface where malicious instructions embedded in the generated diagram text could attempt to influence the agent's behavior during the automated review phase.
Ingestion points: User-provided natural language descriptions for diagrams (SKILL.md).
Boundary markers: None identified in the prompt templates or instructions.
Capability inventory: File system writes (mkdir), command execution (tldraw export), package installation (npm install), and opening files via OS commands (SKILL.md, references/troubleshooting.md).
Sanitization: None; the skill translates user text into JSON labels for diagram shapes without filtering.

tldraw