Judge Runner — Design agent for Claude Code
melta shadow judge の 1 trial を実行する.
How to install Judge Runner
Installs to ~/.claude/agents/tsubotax-melta-ui-judge-runner.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/tsubotax/melta-ui/HEAD/.claude/agents/judge-runner.md -o ~/.claude/agents/tsubotax-melta-ui-judge-runner.md Restart Claude Code, or start a new session, for it to be picked up.
What Judge Runner does
name: judge-runner description: melta shadow judge の 1 trial を実行する。design/judge/run.ts --provider file --phase prepare が書いた tasks/.task.md を渡すと、指定された input.json だけを読み、JSON だけを outputs/.output.txt に書く。判定の中身には立ち入らず、渡された system の規律に従うだけ。 tools: Read, Write
あなたは melta shadow judge の **実行者** です。渡された 1 trial だけを処理します。
手順
- 呼び出し側が指定した
input.jsonを Read する - その JSON の
systemを自分への指示、promptを審査対象の入力として扱う - 応答本文を JSON だけにして、呼び出し側が指定した
outputs/.output.txtに Wr
Alternatives in Design
- Codemod Runner — Writes and runs codemod scripts that replace hardcoded visual values with token references in ui/src/index.css 79.4k ★
- Role Design Author — Generates multi-file role analysis for a brainstorm session — analysis.md index + per-feature files + optional 530 ★
- Asset Producer — Produces one assigned visual asset production unit for the asset stage 517 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Terraform Expert
Use this agent for all Terraform-related tasks: writing or reviewing Terraform code, debugging plan/apply erro
External Judge
Independent sameness scorer for generated UI. Reads TSX output only — no access to the kern skill, the design-
Planning Agent
Design detailed implementation plans for tasks. Create comprehensive step-by-step plans and output as structur
Forge Light
forge plan-execution worker for trivial-tier tasks — config changes, renames, mechanical one-liners with no de
Babbage
Technical architect for mob programming tasks. Invoke at the start of a coding task to produce a design docume
Evolve Architecture Design
Architecture-design agent for the Evolve Loop (Plan archetype). The advisor SELECTS this phase for large or no
Related Skills
Plan Check
You are now a task list completeness checker. Run hard-gate checks against the input tasks.json and output whe
Cha Judge
Adjudicate second-eye claims by running them. Every verdict needs a command, an input, and an output — never a
Loop.Build
Builder/judge protocol: create new tasks, continue existing ones, or advance phases. Usage: /loop.build new ,