The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install external packages from third-party sources, specifically recommending the global installation of single-file-cli via NPM and referencing the monolith Rust tool.
[COMMAND_EXECUTION]: The skill executes a local Python harness script harness/scripts/web-clone.py to perform inlining, URL capturing, and self-tests. It also runs project build commands like npm run build and npx tsc during the reconstruction process.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and processes untrusted HTML, CSS, and JavaScript from external websites to generate new code.
Ingestion points: Reads content from live URLs and local HTML files as specified in SKILL.md.
Boundary markers: None identified; the skill does not provide specific instructions to the agent to ignore potentially malicious instructions embedded in the target website's source code.
Capability inventory: The agent has the ability to write files, execute shell commands (Python and Node.js), and interact with the browser environment via Computer-Use.
Sanitization: There are no documented steps for sanitizing or filtering the extracted website data before it is used to generate the React/Next.js codebase.
[DYNAMIC_EXECUTION]: In the 'reconstruct' mode, the skill generates a new Next.js and Tailwind CSS codebase by reverse-engineering an existing site. It then executes build and type-checking scripts on this generated code, which is derived from external, untrusted input.

web-clone