Agent
Audits existing pages for purposeful illustration and image opportunities, defines one coherent art direction, and produces generation-ready asset plans.
Agent
Audits existing pages for purposeful illustration and image opportunities, defines one coherent art direction, and produces generation-ready asset plans.
Agent
Independently audits and specifies typography, color, spacing, composition, shape, imagery, and control polish for a product-specific frontend redesign.
Agent
Exercises the real rendered interface across tasks, states, viewports, themes, console/network behavior, and before/after evidence.
Command
Create or improve a focused agent definition. Use when a new role is needed. Output: a registered agents/ .md.
Command
Inspect the repository with AI reasoning and build an evidence-backed context pack under .agent/context/. Use after setup or major architecture changes. Output: routing, architecture, modules, flows, data, commands, tests, and risks.
Command
Summarize large context without losing critical facts. Use when output is noisy or near a context limit. Output: a compact, fact-preserving summary.
Command
Turn an ambiguous request into an evidence-backed recommendation via discovery, adaptive specialist review, one challenge round, and chair synthesis. Use for high-uncertainty or high-impact decisions. Output: a decision record; stops for approval, then hands to /spec.
Command
Reproduce, isolate, and fix a failure through a bounded root-cause loop. Use for bugs, crashes, flaky tests, regressions, or incorrect behavior. Output: a verified minimal fix or an evidence-backed blocked report.
Command
Fix only the blocking findings from a review, with no unrelated work. Use after /review. Output: a minimal diff that clears blockers with green gates.
Command
Design and implement a new frontend surface through one intentional, design-system-aware, responsive, accessible, browser-validated pass.
Command
Implement an approved frontend-redesign spec with one coherent writer, browser comparison, and independent visual, UX, interaction, accessibility, and validation specialists.
Command
Implement an approved spec with the smallest reasonable diff, tests, and passing gates. Use after approval. Output: a focused diff and a gate summary.
Command
Audit and improve an existing interface through one evidence-backed, design-system-aware implementation and browser-verification pass.
Command
Interpret an open-ended request and route it to the right HarnessKit command, loop, or specialized agent.
Command
Explain or change behavior-pack configuration. Use to inspect or toggle packs. Output: updated pack state and regenerated exports.
Command
Audit an existing web page for purposeful image and illustration opportunities, then write an approval-gated three-tier asset spec with one coherent style and generation-ready prompts, dimensions, and delivery metadata.
Command
Review the diff against the spec for correctness, regressions, and missing tests. Use after implementation. Output: labelled findings and a verdict.
Command
Create a highly detailed specification through deeper discovery and a larger, structured question batch before approval.
Command
Produce an approval-gated, browser-evidenced redesign specification for an existing frontend surface using independent UX, visual, interaction, design-system, accessibility, and validation specialists.
Command
Adaptively create a reviewable specification: inspect context, ask minimal blockers, consult selected specialists when useful, then wait for approval.
Command
Explain an existing specification in concise plain language before approval. Use after /spec or whenever a developer needs to understand a spec quickly.
Command
Run an actionable UX and accessibility review of user-facing changes. Use for UI work. Output: prioritized blocking issues and polish.
Skill Claude CodeCodex
Build and verify semantic, keyboard-operable, perceivable, and recoverable interfaces.
Skill Claude CodeCodex
Create a focused agent definition.