The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied workflow descriptions and interpolates them directly into an HTML template to generate a diagram. There is no explicit instruction to sanitize or escape this input. If a user provides a description containing HTML or JavaScript, it would be included in the generated HTML and rendered by a headless browser.
Ingestion points: Workflow descriptions are collected from the user as described in Phase 1 of SKILL.md.
Boundary markers: No explicit delimiters or warnings (e.g., 'ignore embedded instructions') are implemented when inserting user data into the HTML structure in Phase 3 of SKILL.md.
Capability inventory: The screenshot-diagram.js script launches a Chromium browser and executes page.goto on the generated file, creating a surface for script execution.
Sanitization: Instructions in SKILL.md for parsing and HTML generation do not mention sanitizing or validating user-provided strings before inclusion in the DOM.
[EXTERNAL_DOWNLOADS]: The skill requires downloading the playwright Node.js package and installing the Chromium browser binary.
Evidence: package.json lists playwright as a dependency. The setup script runs npx playwright install chromium. These are legitimate requirements for the skill's stated purpose of capturing screenshots.
[COMMAND_EXECUTION]: The skill executes a local Node.js script with arguments derived from the diagram name.
Evidence: SKILL.md instructs the agent to execute node screenshot-diagram.js <diagram-name>. While the agent typically generates the name, a lack of strict validation on the diagram-name could theoretically lead to command injection if the agent is manipulated into using shell metacharacters.

create-workflow-diagram