The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to run internal utility scripts located in .claude/skills/impeccable/scripts/. These scripts handle project context gathering (context.mjs), design seeding (concept-seed.mjs), and the live iteration server (live-server.mjs).
[EXTERNAL_DOWNLOADS]: The skill performs network requests to several destinations:
https://impeccable.style/api/roll: Fetches design seeds and challengers.
https://api.openai.com/v1/images/generations: Used in generate-image.mjs to create UI mocks if the user has provided an OPENAI_API_KEY.
https://impeccable.style/api/version: Checks for skill updates during the boot process.
https://fonts.googleapis.com: Referenced in serve-question.mjs to load typography for the visual decision page.
[INDIRECT_PROMPT_INJECTION]: The 'Live Mode' functionality (live.mjs, live-poll.mjs) creates an attack surface by allowing the agent to process UI edits stashed in the browser. However, the skill explicitly mitigates this in live-copy-edit-agent.mjs by instructing the agent to treat input as literal data and valid source syntax only, maintaining a strict boundary between user copy and agent instructions.
[PROMPT_INJECTION]: A static analysis hint flagged reference/craft.md for concealment. Upon review, this is a benign instruction related to the deprecation of an alias (craft), advising the agent to use natural language instead of forced command names to improve the user experience.

impeccable