The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill transforms user-provided requirements into functional code prototypes (React, Node.js, Python, HTML) without defining clear safety boundaries or implementing sanitization logic for the input.
Ingestion points: User prompts describing features or logic for an "MVP", "prototype", or "proof of concept".
Boundary markers: The skill instructions do not provide the agent with delimiters or specific directives to ignore potentially malicious instructions embedded within the user's requirements.
Capability inventory: The skill generates files on the local filesystem and provides explicit instructions for the user to run the resulting code using shell commands like npm install and npm run dev.
Sanitization: There is no evidence of logic to validate or escape user-provided requirements before they are interpolated into the generated script templates.
[COMMAND_EXECUTION]: The skill provides templates and instructions that direct users to execute shell commands, including package installations (npm install) and starting development servers (npm run dev), on code generated by the AI.
[EXTERNAL_DOWNLOADS]: The single-file HTML prototype template includes a script tag that fetches the Vue.js library from unpkg.com, which is a well-known content delivery network for front-end assets.
[METADATA_POISONING]: The SKILL.json file contains hardcoded absolute paths referencing a specific user's home directory (/home/toowired/) in the Next Steps if Validated and Quick Reference sections, suggesting the configuration may have been exported from a specific local environment.

rapid-prototyper