The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes bundled Node.js scripts (scripts/breadboard.mjs and scripts/serve.mjs) to perform its core functions of rendering maps and serving them locally.
The scripts use internal boundary checks to confine file access to the target repository root.
The serve.mjs script starts a local HTTP server that is strictly limited to the loopback address (127.0.0.1).
[INDIRECT_PROMPT_INJECTION]: The skill processes JSON data (breadboard.json) which could theoretically contain malicious instructions if the file is sourced from an untrusted collaborator.
Ingestion points: The breadboard.json file is read by scripts/breadboard.mjs and processed by scripts/renderer.mjs.
Boundary markers: The instructions guide the agent to verify behavior against existing code and documentation before updating the map.
Capability inventory: The skill can write files (the generated HTML artifacts) and start a local HTTP server.
Sanitization: scripts/renderer.mjs implements an HTML sanitizer that explicitly blocks <script>, <style>, <iframe>, and event handlers (e.g., onmouseover) within the JSON content to prevent cross-site scripting (XSS) in the generated viewer.
[SAFE]: The skill implements an integrity verification system. It checks the SHA-256 hashes of its own bundled assets against a provenance.json file before execution to ensure the skill has not been tampered with.

breadboard