The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external web content from arbitrary URLs or local files, creating an attack surface for instructions embedded within the source HTML/JS to influence the agent.
Ingestion points: External websites via URLs or local HTML files (SKILL.md).
Capability inventory: Writing files to the local system, executing build commands (npm, tsc), and performing network operations via capture engines.
Boundary markers: There are no explicit instructions for the agent to sanitize or ignore potential instructional content within the source data being cloned.
Sanitization: The skill relies on AI reverse-engineering without specific filtering for malicious prompt patterns in the input source code.
[COMMAND_EXECUTION]: The skill utilizes shell commands to perform cloning, testing, and building operations.
Execution of local scripts: Runs python3 harness/scripts/web-clone.py for core functionality.
Build toolchain: Invokes npx tsc --noEmit and npm run build to validate the fidelity of reconstructed codebases.
Tool management: Recommends the global installation of single-file-cli to enable live capture features.
[EXTERNAL_DOWNLOADS]: The skill's primary purpose involves fetching data from external network locations.
Fetches content from user-specified URLs to create snapshots or reconstructed codebases.
References external repositories on GitHub, including gildas-lormeau/SingleFile and JCodesMore/ai-website-cloner-template, as technical foundations for the cloning process.

web-clone