The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill requires the agent to author self-contained HTML files containing executable inline JavaScript and then interact with these files using browser automation skills such as /agent-browser or claude-in-chrome to verify functionality. This cycle of generating and executing local code creates a sandbox-escape risk depending on the capabilities of the browser tools used.
[INDIRECT_PROMPT_INJECTION]: The skill generates prototype code based on requirements and "unknowns" derived from the user's request or existing work context. This creates a surface where external or untrusted data can influence the behavior of the scripts executed by the agent during the testing phase.
Ingestion points: Prototype requirements and logic are derived from inputs passed to the skill or the agent's current workspace (Step 1).
Boundary markers: The instructions do not specify the use of delimiters or safety headers to prevent the agent from obeying malicious instructions embedded within the processed workspace data.
Capability inventory: The skill uses file system write access to save HTML files to the .turbo/prototypes/ directory and utilizes browser automation tools to load and execute these local files.
Sanitization: While the skill provides robust logic to sanitize the output filename (slug) by removing non-alphanumeric characters and trimming length, it lacks specific sanitization, escaping, or validation protocols for the content of the generated inline scripts within the HTML file.

prototype