micode

OpenCode plugin with structured Brainstorm → Plan → Implement workflow and session continuity.

micode.mp4

Quick Start

Add to ~/.config/opencode/opencode.json:

{ "plugin": ["micode"] }

Then run /init to generate ARCHITECTURE.md and CODE_STYLE.md.

Workflow

Brainstorm → Plan → Implement
     ↓         ↓        ↓
  research  research  executor

Brainstorm

Refine ideas into designs through collaborative questioning. Fires research subagents in parallel. Output: thoughts/shared/designs/YYYY-MM-DD-{topic}-design.md

Plan

Transform designs into implementation plans with bite-sized tasks (2-5 min each), exact file paths, and TDD workflow. Output: thoughts/shared/plans/YYYY-MM-DD-{topic}.md

Implement

Execute in git worktree for isolation. The Executor orchestrates implementer→reviewer cycles with parallel execution via fire-and-check pattern.

Session Continuity

Maintain context across sessions with structured compaction. Run /ledger to create/update thoughts/ledgers/CONTINUITY_{session}.md.

Commands

Command	Description
/init	Initialize project docs
/ledger	Create/update continuity ledger
/search	Search past plans and ledgers

Agents

Agent	Purpose
commander	Orchestrator
brainstormer	Design exploration
planner	Implementation plans
executor	Orchestrate implement→review
implementer	Execute tasks
reviewer	Check correctness
codebase-locator	Find file locations
codebase-analyzer	Deep code analysis
pattern-finder	Find existing patterns
project-initializer	Generate project docs
ledger-creator	Continuity ledgers
artifact-searcher	Search past work

Tools

Tool	Description
ast_grep_search	AST-aware code pattern search
ast_grep_replace	AST-aware code pattern replacement
look_at	Extract file structure
artifact_search	Search past plans/ledgers
btca_ask	Query library source code
pty_spawn	Start background terminal session
pty_write	Send input to PTY
pty_read	Read PTY output
pty_list	List PTY sessions
pty_kill	Terminate PTY

Hooks

• Think Mode - Keywords like "think hard" enable 128k token thinking budget
• Ledger Loader - Injects continuity ledger into system prompt
• Auto-Compact - At 50% context usage, automatically summarizes session to reduce context
• File Ops Tracker - Tracks read/write/edit for deterministic logging
• Artifact Auto-Index - Indexes artifacts in thoughts/ directories
• Context Injector - Injects ARCHITECTURE.md, CODE_STYLE.md
• Token-Aware Truncation - Truncates large tool outputs

Configuration

Model Configuration

micode reads your default model from opencode.json:

{
  "model": "github-copilot/gpt-5-mini",
  "plugin": ["micode"]
}

All micode agents will use this model automatically.

micode.json

Create ~/.config/opencode/micode.json for micode-specific settings:

{
  "agents": {
    "brainstormer": { "model": "openai/gpt-4o", "temperature": 0.8 },
    "commander": {
      "maxTokens": 8192,
      // Configure reasoning effort per agent
      "thinking": { "type": "enabled", "budgetTokens": 100000 }
    }
  },
  "features": {
    "mindmodelInjection": true,
    "context7": false
  },
  "compactionThreshold": 0.5,
  "fragments": {
    "commander": ["custom-instructions.md"]
  }
}

Note: Both .json and .jsonc formats are supported. JSONC allows comments and trailing commas.

Options

Option	Type	Description
agents	object	Per-agent overrides (model, temperature, maxTokens, thinking)
features.mindmodelInjection	boolean	Enable mindmodel context injection
features.context7	boolean	Register the bundled context7 MCP server. Set false when it already comes from another tool or proxy. Default: true
compactionThreshold	number	Context usage threshold (0-1) for auto-compaction. Default: 0.7
fragments	object	Additional prompt fragments per agent

Model Resolution Priority

1. Per-agent override in micode.json (highest)
2. Default model from opencode.json "model" field
3. Plugin default (fallback)

Model Syntax

Models use provider/model format. The provider must match exactly what's in your opencode.json:

{
  "provider": {
    "github-copilot": {
      "models": { "gpt-5-mini": {} }
    }
  }
}

Use "model": "github-copilot/gpt-5-mini" (not github/copilot:gpt-5-mini).

Development

git clone git@github.com:vtemian/micode.git ~/.micode
cd ~/.micode && bun install && bun run build

// Use local path
{ "plugin": ["~/.micode"] }

Release

npm version patch  # or minor, major
git push --follow-tags

Philosophy

1. Brainstorm first - Refine ideas before coding
2. Research before implementing - Understand the codebase
3. Plan with human buy-in - Get approval before coding
4. Parallel investigation - Spawn multiple subagents
5. Isolated implementation - Use git worktrees
6. Continuous verification - Implementer + Reviewer per task
7. Session continuity - Never lose context

micode vs oh-my-opencode

Both are OpenCode plugins, but with different philosophies:

Aspect	micode	oh-my-opencode
Philosophy	Opinionated workflow (brainstorm→plan→implement)	Batteries-included framework
Agent Design	Role-based (Brainstormer, Planner, Executor)	Greek mythology theme (Sisyphus, Atlas, Prometheus)
Parallelism	Batch-first: 10-20 concurrent micro-tasks (2-5 min each)	Background tasks with tmux visual monitoring
Code Guidance	Mindmodel system with project-specific patterns	Comment checker, keyword modes (ultrawork)
Context Recovery	Ledger system (CONTINUITY files)	AGENTS.md hierarchy, preemptive compaction
Workflow	TDD-enforced with adaptation over escalation	Category-based delegation (visual-engineering, ultrabrain)
Configuration	Focused options	Extensive (34 hooks, 11 agents, fallback chains)

When to Choose micode

• You want a structured brainstorm→plan→implement workflow
• You prefer TDD-driven implementation with test-first development
• You need project-specific pattern enforcement via mindmodel
• You want high parallelism on granular tasks (10-20 concurrent micro-tasks)
• You value session continuity via structured ledgers

When to Choose oh-my-opencode

• You want maximum flexibility and configuration options
• You prefer keyword-driven modes (e.g., "ultrawork", "analyze")
• You need extensive model fallback chains with subscription detection
• You like category-based task delegation (visual-engineering, infrastructure)
• You want visual monitoring via tmux integration

Inspiration

• oh-my-opencode - Plugin architecture
• HumanLayer ACE-FCA - Structured workflows
• Factory.ai - Structured compaction research