The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The critique command invokes npx impeccable, which fetches and executes a package from the public NPM registry to perform automated anti-pattern detection.
[COMMAND_EXECUTION]: Multiple utility scripts (live.mjs, live-server.mjs, pin.mjs) use Node.js child_process modules to spawn background processes, manage a local design relay server, and execute shell-based commands for environment management.
[COMMAND_EXECUTION]: The pin command programmatically modifies the agent's environment by creating new skill directories and SKILL.md files in harness folders (e.g., .claude/skills/) to establish permanent shortcuts for sub-commands.
[DATA_EXFILTRATION]: The skill starts a local HTTP helper server with a /source endpoint that allows reading content from the project directory. While this is used for design iteration and is protected by a UUID authorization token and directory traversal checks, it provides a functional surface for local file exposure via the network stack.
[COMMAND_EXECUTION]: Instructions in live.md guide the agent to perform extensive source code modifications, including element wrapping and patching the project's Content Security Policy (CSP) to allow connections to the local helper server.

impeccable