critique-loop CLAUDE.md

A Claude Code plugin for structured, turn-by-turn discussions between two spawned coding agents, with the main agent coordinating the exchange.

In plain words
What is it for?
Use it for planning sessions, code reviews, and pair-programming conversations where one agent critiques another.
Why use it?
It provides an organized way to challenge plans, review code, or compare ideas without mixing the discussion into the main terminal output.

Instructions file

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add instructions/jarfa/critique-loop/claude-md
Clone the repo
git clone --depth 1 https://github.com/jarfa/critique-loop
Per session 1,041 This file is loaded in full into every session.
When invoked 1,041 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.01041 $0.01041
Opus 5 $0.00521 $0.00521
Sonnet 5 $0.00208 $0.00208
Haiku 4.5 $0.00104 $0.00104

Measured 2d ago against content hash 4a0f4811f83f, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

critique-loop CLAUDE.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

CLAUDE.md · 93 lines

How it starts

The opening of the file, as written. The whole thing — 93 lines — stays where its author put it; the contents beside it link to each section on GitHub.

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project Overview

Critique Loop is a Claude Code plugin that enables structured, turn-based dialogues between two spawned subagents. The user's Claude instance orchestrates but never writes dialogue turns—keeping all dialogue output hidden from the terminal. Designed for planning sessions, code reviews, and pair programming where automated critique is valuable.

Status: Implementation complete.

Architecture

Core Components

  1. Orchestrator Skill (plugins/critique-loop/skills/critique-loop/skill.md) - Coordinates the dialogue: gathers user context, creates dialogue file header, spawns two subagents, manages turn-taking via Task tool resume, reports outcomes. Never writes dialogue turns directly.

  2. Participant Instructions (plugins/critique-loop/dialogue-partner-instructions.md) - Standalone protocol rules both subagents follow. Kept separate from orchestrator to ensure unbiased dialogue.

  3. Dialogue Files (.dialogues/<YYYYMMDD-HHMMSS>-<topic-slug>.md) - Append-only markdown files storing the conversation. Format:

    # Dialogue: <topic>
    Started: YYYY-MM-DD HH:MM
    Max rounds: 5
    
    Participants:
      - <role1> @ <directory>
      - <role2> (subagent)            # when partner is Claude (default)
      - <role2> (codex)               # when partner is Codex
    
    ---
    
    ## [<role>] Round N — YYYY-MM-DD HH:MM
    <message>
    Status: AWAITING <other-role>|PROPOSING_DONE|DONE|STUCK
    

Coordination Model

  • Two-subagent architecture: Orchestrator spawns Subagent A (user's role) and Subagent B (partner role); both write turns in subagent context (hidden from terminal)
  • Partner selection: Subagent B defaults to Claude but can be Codex CLI. Codex is stateless — each turn is a fresh codex exec --full-auto --ephemeral invocation. It reads the dialogue file for context and appends via heredoc (cat >>), since it lacks a Write tool. Claude subagents use the Task tool with resume for multi-turn context.
  • Task tool orchestration: Orchestrator spawns, resumes, and coordinates subagents; subagents return after writing each turn
  • Append-only protocol: Prevents conflicting edits; turns are immutable
  • Status-driven flow: Last line of each turn declares who acts next
  • Unbiased partner: Subagent B receives no user context—reads dialogue file fresh for objective critique
  • Codex failure modes: Non-zero exit code, file not modified, malformed turn (invalid Status: line), and timeout (5-minute cap). Each offers the user options: retry, switch to Claude partner, or end dialogue.

Read the full file on GitHub · 93 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 93 lines · 1,041 tokens per session scan A 4a0f4811f83f

Subscribe to this mod's changes

critique-loop CLAUDE.md is an instructions file published in the GitHub repository jarfa/critique-loop (5 stars, last pushed 5mo ago), licensed MIT. It adds 1,041 tokens to every session, about $0.0052 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.