The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Node.js scripts (mirror-site.mjs and compare-site.mjs) located within the skill's root directory. These scripts perform the core mirroring and auditing tasks as intended by the skill's design.
[EXTERNAL_DOWNLOADS]: During its first execution, the skill uses npx to retrieve esbuild. This is a well-known and industry-standard JavaScript bundling tool used here to ensure that mirrored scripts can function correctly in a local file environment.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests data from external websites specified by the user during crawling and comparison operations.
Boundary markers: The AI agent does not directly interpret the raw content of the websites; it only reads high-level summaries and audit logs generated by the execution of the tools.
Capability inventory: The skill uses the Node.js runtime to execute scripts, writes files to the user's ~/Downloads directory, and performs network GET requests to external hosts.
Sanitization: Results are sanitized by the scripts, which produce structured, one-line-per-topic summaries for the agent. This prevents raw HTML or script content from being directly processed by the LLM, mitigating the risk of instructions embedded in websites influencing agent behavior.

saving-websites-locally