~ / rigs / raiki61 / claude-plugins

raiki61/claude-plugins

Two Claude Code loops that keep reviewing and researching until they reach a fixed point: the implementer and the grader are held in separate contexts, and the machine may only ever declare failure, never convergence. Japanese only for now.

↗ GitHub ★ 1 MIT updated today project Claude Code Claude plugin
share on X
ARCHETYPE
Orchestrator
A bench of specialised subagents. The main agent mostly delegates.
CONTEXT TAX · EVERY TURN
~1.2k tokens
Featherweight · median rig: 2.3k · breakdown
GUARDRAILS
1/5
No YOLO mode · details

Copy this rig

# review before running: this installs third-party code
$ npx degit raiki61/claude-plugins/.claude ./rig-claude-plugins  # inspect, then merge into .claude/

MCP servers are added to Claude Code at local scope; env vars are shown as YOUR_… placeholders — we never store values. Files are fetched with degit into a separate folder so you can review before merging.

$ npx degit raiki61/claude-plugins/gates/skills/coldread .claude/skills/coldread
$ npx degit raiki61/claude-plugins/works/.shared/borrow/superpowers/6.4.2/skills/receiving-code-review .claude/skills/receiving-code-review
$ npx degit raiki61/claude-plugins/works/.shared/borrow/superpowers/6.4.2/skills/requesting-code-review .claude/skills/requesting-code-review
$ npx degit raiki61/claude-plugins/works/.shared/borrow/superpowers/6.4.2/skills/systematic-debugging .claude/skills/systematic-debugging
$ npx degit raiki61/claude-plugins/works/.shared/borrow/superpowers/6.4.2/skills/test-driven-development .claude/skills/test-driven-development
$ npx degit raiki61/claude-plugins/works/.shared/borrow/superpowers/6.4.2/skills/verification-before-completion .claude/skills/verification-before-completion
$ npx degit raiki61/claude-plugins/works/skills/works .claude/skills/works
$ curl -fsSL --create-dirs -o .claude/agents/blind-judge.md https://raw.githubusercontent.com/raiki61/claude-plugins/main/agents/blind-judge.md
$ curl -fsSL --create-dirs -o .claude/agents/cold-reader.md https://raw.githubusercontent.com/raiki61/claude-plugins/main/agents/cold-reader.md
$ curl -fsSL --create-dirs -o .claude/agents/inspector.md https://raw.githubusercontent.com/raiki61/claude-plugins/main/agents/inspector.md
$ curl -fsSL --create-dirs -o .claude/agents/investigator.md https://raw.githubusercontent.com/raiki61/claude-plugins/main/agents/investigator.md
$ curl -fsSL --create-dirs -o .claude/agents/judge.md https://raw.githubusercontent.com/raiki61/claude-plugins/main/agents/judge.md
$ curl -fsSL --create-dirs -o .claude/agents/reader.md https://raw.githubusercontent.com/raiki61/claude-plugins/main/agents/reader.md

Merge into .claude/settings.json (project) or ~/.claude/settings.json (user). Hook commands reference scripts in the source repo — copy those too.

{
  "permissions": {
    "deny": [
      "Bash(bash tests/run.sh *)",
      "Bash(bash graphloops/tests/run.sh *)",
      "Bash(sh tests/run.sh *)",
      "Bash(sh graphloops/tests/run.sh *)",
      "Bash(./tests/run.sh *)",
      "Bash(./graphloops/tests/run.sh *)"
    ]
  }
}

Skills (7)

Subagents (6)

blind-judge
model: opus
内部の実態を遮断した独立導出・突合の目。道具を一切持たず、渡された前提・問いだけから考える上位の役。
cold-reader
model: sonnet
文脈を遮断した初見の読み手。道具を一切持たず、プロンプトに貼られた本文だけを読む。
inspector
model: sonnet
呼び出し側と別 context で採点する読み取り専用の目。リポジトリは読めるが書けず、shell も web も持たない。
investigator
model: sonnet
リポジトリの履歴と外(git・gh・web)まで探しに行く読み取り役。shell は持つが、変更は禁止。
judge
model: opus
呼び出し側が覆せない判定を下す上位の目。リポジトリと web は読めるが書けず、shell を持たない。
reader
model: sonnet
文書を知らない読み役。用事だけを渡され、リポジトリの入口から自分で探して読む。firstread-loop 専用。

Slash commands (4)

/doctor-loop/firstread-loop/research-loop/review-loop

Permissions

deny (6)
Bash(bash tests/run.sh *)
Bash(bash graphloops/tests/run.sh *)
Bash(sh tests/run.sh *)
Bash(sh graphloops/tests/run.sh *)
Bash(./tests/run.sh *)
Bash(./graphloops/tests/run.sh *)
ask (0)
—
allow (0)
—

Similar rigs

copied ✓