companion

Polishes and improves the UI in your project. Covers concentric border radius, optical alignment, surface depth, contextual icons, hit areas and more.

Turn technical conversations, bugs, CI failures, review comments, and repeated agent mistakes into durable project improvements. Use when the user asks what should be automated, documented, tested, added to AGENTS.md/CLAUDE.md, wired into CI, or changed so the same issue does not happen again. This skill is for both non-technical and technical users: it explains the learning in plain English, inspects the repository, chooses the right prevention layer, proposes changes first, and only edits project files when explicitly asked.

Clean Impeccable frontend design skill. Use when the user wants to design, redesign, critique, audit, polish, harden, optimize, adapt, clarify, colorize, animate, or otherwise improve a frontend interface. Covers websites, landing pages, dashboards, product UI, app shells, components, forms, onboarding, empty states, accessibility, responsive behavior, visual hierarchy, UX copy, motion, design systems, anti-pattern detection, and AI-looking UI cleanup. This clean Companion edition is agent-only: no CLI, no hooks, no live mode, no install commands, no external commercial workflow.

This skill encodes Emil Kowalski's philosophy on UI polish, component design, animation decisions, and the invisible details that make software feel great.

Grill the user relentlessly about a plan, decision, or idea. Use when the user wants to stress-test their thinking, or uses any 'grill' trigger phrases.

Turn a prompt, bug report, issue ID, or ticket URL into a repository-grounded implementation plan for human approval. Use when the user asks to plan a PR, prepare a ticket for implementation, work out how to build a feature, or set up the plan-pr workflow. Reads tracker context and repo rules, routes to a fixed set of design and TDD skills, and defines observable acceptance criteria. After approval the same agent implements, then invokes ship-pr-dev. For shipping an existing branch or fixing PR CI without a planning request, use ship-pr-dev directly.

Verify that the compiled Companion can load a skill and execute its packaged shell resource.

Delegation mode for open-code-review (OCR). Instead of OCR calling an LLM endpoint, this skill instructs the host agent to perform the code review itself, using OCR only for deterministic engineering: file selection and rule resolution. Use when the host agent should drive the review with its own LLM capabilities.

Ship PR: credit-aware autonomous pull request readiness workflow. Use when the user asks to ship work, prepare a branch for review, create or update a PR, make CI green, clean up a branch before PR, or produce a PR a human can confidently merge. This skill may edit code, run checks, commit, push, create/update a PR, and iterate on CI, but it keeps Git ownership in the coordinator, delegates bounded work to cheaper host-native workers, uses review-code-dev v2 (Alibaba OCR delegation) as the only independent read-only review gate, waits for all visible latest-SHA CI to be green, runs capture-learning-tools report-only, and never merges.

Test-driven development. Use when the user wants to build features or fix bugs test-first, mentions "red-green-refactor", or wants integration tests.