Humanize for OpenCode

Version: 1.0.0

Iterative development with independent AI review. Build with confidence through continuous feedback loops using GLM-5.1 for code review.

What is RLCR?

RLCR stands for Ralph-Loop with Code Review - an iterative development cycle where AI-generated code is continuously refined through independent review feedback.

Core Concepts

• Iteration over Perfection - Continuous feedback loops catch issues early
• One Build + One Review - OpenCode implements, GLM-5.1 independently reviews
• Begin with the End in Mind - Verify you understand the plan before execution

Quick Start

1. Install

# Clone to OpenCode packages directory
git clone https://github.com/humanize-org/humanize-opencode.git \
  ~/.cache/opencode/packages/humanize

# Set permissions
chmod +x ~/.cache/opencode/packages/humanize/scripts/*.sh
chmod +x ~/.cache/opencode/packages/humanize/hooks/*.sh

# Setup monitoring helper
echo 'source ~/.cache/opencode/packages/humanize/scripts/humanize.sh' >> ~/.bashrc
source ~/.bashrc

2. Configure

# Create user config
mkdir -p ~/.config/humanize
cat > ~/.config/humanize/config.json << 'EOF'
{
  "reviewer_model": "GLM-5.1",
  "reviewer_effort": "high",
  "quiz_model": "GLM-5.1",
  "monitoring_model": "GLM-5",
  "max_iterations": 42
}
EOF

# Set API key
export ZHIPU_API_KEY="your-api-key"

3. Generate a Plan

# From draft to structured plan
/skill:gen-plan --input feature-draft.md --output plan.md

4. Run the Loop

# Start RLCR loop
/skill:humanize-rlcr plan.md

# Or with options
/skill:humanize-rlcr plan.md --max 20 --yolo

5. Monitor Progress

# Watch loop progress
humanize monitor rlcr

# Quick status
humanize status

Features

Plan Understanding Quiz

Verifies you understand the plan before execution to prevent "wishful coding".

Goal Tracking System

Tracks acceptance criteria, completed items, and plan evolution.

Independent Code Review

GLM-5.1 provides objective code review with [P0-9] severity markers.

Agent Teams Mode

Parallelize independent tasks with --agent-teams flag.

Real-time Monitoring

Track progress with humanize monitor command.

Commands

Command	Purpose
/skill:humanize-rlcr plan.md	Start iterative development loop
/skill:gen-plan --input draft.md --output plan.md	Generate structured plan
/skill:ask-model "question"	One-shot model consultation
humanize monitor rlcr	Monitor loop progress
humanize status	Quick status check
humanize doctor	Run diagnostics

Configuration

4-layer hierarchy (lowest to highest priority):

1. Plugin defaults: ~/.cache/opencode/packages/humanize/config/default_config.json
2. User config: ~/.config/humanize/config.json
3. Project config: .humanize/config.json
4. CLI flags

Prerequisites

• OpenCode CLI >= 1.0.0
• Git >= 2.20
• Zhipu GLM API access or local deployment
• jq (for JSON parsing)

Directory Structure

~/.cache/opencode/packages/humanize/
├── skills/          # OpenCode skills
├── scripts/         # Shell scripts
├── hooks/           # Review hooks
├── templates/       # Plan/summary templates
├── config/          # Default configuration
└── agents/          # Agent instructions

Project-level state in <project>/.humanize/:

• rlcr/<timestamp>/ - Loop data
• config.json - Project config

Troubleshooting

Issue: GLM-5.1 API timeout

Solution: Check network connection, increase --timeout value.

Issue: ZHIPU_API_KEY not set

Solution: export ZHIPU_API_KEY="your-api-key"

Issue: Git clean check failed

Solution: Commit or stash current changes.

Issue: State file corrupted

Solution: Run humanize doctor to diagnose.

Documentation

• Design Specification
• Usage Guide

License

MIT

Credits

Derived from humanize for Claude Code.