Skip to content

zackbart/motif

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

17 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Motif

A structured 5-stage development workflow for AI coding agents: Research → Plan → Scaffold → Build → Validate.

Works with any agent that supports the Agent Skills standard — including Claude Code, OpenAI Codex CLI, Cursor, and 40+ other platforms.

Motif ensures you research before coding, plan with tradeoff analysis, track tasks explicitly, and validate against the original goal.

Installation

Any Agent (via npx skills)

npx skills add zackbart/motif

This installs skills to whichever agent platforms you have installed.

Claude Code (full plugin)

claude plugin add zackbart/motif

Includes all skills plus dedicated subagents (researcher, critic, docs-scanner, docs-writer) and session hooks.

Codex CLI

git clone https://github.com/zackbart/motif.git ~/.agents/skills/motif

Usage

Invoke skills using your agent's skill invocation method:

/motif:dev <task description>        # Claude Code
/motif:optimize-prompt <prompt goal> # Claude Code
/motif:second-opinion                # Claude Code
/motif:update-docs [focus area]      # Claude Code
$dev <task description>              # Codex CLI

Or use werk for persistent task tracking:

/motif:dev werk <task description>   # Claude Code
$dev werk <task description>         # Codex CLI

Workflow Stages

Each stage requires your approval before advancing to the next.

1. Research

Performs deep, read-only codebase exploration. Depth scales automatically based on task complexity (light / medium / heavy). On Claude Code, this can be delegated to a dedicated research subagent for efficiency.

2. Plan

Develops an implementation approach based on the research findings. Medium and heavy tasks include tradeoff analysis and risk assessment.

3. Scaffold

Creates a structured task list from the plan using the platform's available task tracking tools, or the werk CLI.

4. Build

Executes the plan by working through tasks in order. Stops and reports if blockers are encountered.

5. Validate

Verifies the implementation against the original request: goal check, test verification, diff review, and loose-end identification.

Claude Code Extras

When installed as a Claude Code plugin, motif includes additional features:

Feature Description
Research subagent Dedicated agent on Sonnet for codebase exploration
Critic subagent Adversarial reviewer for pressure-testing plans
Docs scanner/writer subagents Parallel doc auditing and updating
Session hooks Reminds you about the workflow on session start

Platform Compatibility

Feature Claude Code Codex CLI Other Agents
5-stage workflow Yes Yes Yes
Research subagent Dedicated agent Inline Inline
Task tracking Native task tools update_plan / werk Platform-dependent / werk
Subagents researcher, critic, docs-scanner, docs-writer Inline Inline
Distribution Plugin marketplace Git clone npx skills

Project Structure

skills/dev/              # Core 5-stage workflow (universal skill)
  references/            # Prompt engineering reference guides
skills/optimize-prompt/  # Prompt engineering assistant (universal skill)
skills/second-opinion/   # Adversarial plan critique (universal skill)
skills/update-docs/      # Documentation auditor and updater (universal skill)
agents/                  # Subagents: researcher, critic, docs-scanner, docs-writer
hooks/               # Session lifecycle hooks (Claude Code)
AGENTS.md            # Session priming (Codex CLI)
.claude-plugin/      # Plugin manifest (Claude Code)

License

MIT

About

A Claude Code plugin that encodes a structured 5-stage development workflow — Research, Plan, Scaffold, Build, Validate — with approval gates between each stage. Language and codebase agnostic. Includes a deep research subagent, prompt drafting assistant, and documentation updater.

Resources

Stars

Watchers

Forks

Packages

 
 
 

Contributors