The Agent Skills Directory

[PROMPT_INJECTION]: The skill retrieves and processes data from arbitrary external URLs, creating a surface for indirect prompt injection.
Ingestion points: HTML, CSS, and JavaScript content from user-specified target websites (SKILL.md, Phase 1).
Boundary markers: No explicit boundary markers or instructions to ignore embedded prompts are provided when processing external content.
Capability inventory: The skill utilizes curl for downloads, npm for package installation, fs.writeFileSync for file operations, and playwright-cli for browser automation and JavaScript execution.
Sanitization: There is no evidence of sanitization or filtering of the extracted text or code before it is used to generate the local project or TEARDOWN.md.
[COMMAND_EXECUTION]: The skill uses subprocess calls to execute system commands to facilitate its primary function.
curl is used to download script bundles and assets from target websites to the local output directory (SKILL.md, Phase 1, Step 7).
npm install (or npm i) is used to set up the build environment and install dependencies for the cloned site (SKILL.md, Phase 2).
[EXTERNAL_DOWNLOADS]: The skill identifies and downloads external assets and code based on user-provided target URLs.
It fetches images, fonts, and stylesheets from target URLs (fidelity-fast-path.md, Step 2).
It downloads JavaScript bundles from target sites for analysis and reconstruction (SKILL.md, Phase 1, Step 7).
[REMOTE_CODE_EXECUTION]: The skill executes JavaScript obtained from external websites as part of its analysis and verification process.
playwright-cli is used to execute probe scripts within the context of the remote website to extract runtime parameters (playwright-cli-recipes.md).
The 'shader-extract' track captures and replays GPU effects, which involve executing extracted shader code in a local browser environment for verification (SKILL.md, Phase 3).

clone-site