do-it-agent

Multi-agent orchestrator plugin for Copilot CLI. Takes a prompt through intent analysis, implementation planning, wave-based coding, and quality grading — fully automated from idea to PR.

How It Works

User Prompt
    │
    ▼
┌─────────────────────────────────┐
│       do-it-orchestrator        │
│                                 │
│  Phase 1 → Analyzer             │  Understand intent, explore codebase
│  Phase 2 → Planner              │  Create implementation plan + docs
│  Phase 3 → Plan Reviewer        │  Validate plan (loop if needed)
│  Phase 4a → QA Contract         │  Agree on quality criteria
│  Phase 4b → Implementor         │  Wave-based implementation
│  Phase 5 → Grader               │  Grade against QA contract
│           ↺ (loop 4b–5)         │  Fix issues and re-grade
│  Phase 6 → Finalize             │  Push branch, create PR, check CI
└─────────────────────────────────┘

Agents

Agent	Role
do-it-orchestrator	Pipeline coordinator — manages state and dispatches phases
do-it-analyzer	Explores codebase, parses intent, identifies risks
do-it-planner	Creates implementation plan with feature waves and decision records
do-it-plan-reviewer	Validates plan for completeness, simplicity, and feasibility
do-it-implementor	Negotiates QA criteria and implements code wave-by-wave
do-it-grader	Sets quality standards and grades implementation with structured verdicts

File Handoff

All inter-agent state is managed through the filesystem:

.do-it/runs/<run-id>/
├── state.json              # Pipeline state (phase, iteration, verdicts)
├── analysis/
│   └── context.md          # Intent, codebase map, risks
├── plan/
│   ├── plan.md             # Master implementation plan
│   ├── feature_split.md    # Wave breakdown and order
│   ├── decisions.md        # Architecture decision records
│   ├── gotchas.md          # Edge cases and pitfalls
│   ├── evidence.md         # Code references
│   └── quality_plan.md     # Agreed QA contract
├── reviews/
│   └── plan_review_N.md    # Plan reviewer findings
└── iterations/
    └── N/
        └── grade_report.md # Structured grader verdict

Quick Start — Copilot Plugin

This repository ships as a Copilot CLI / VS Code plugin (docs).

File	Purpose
plugin.json	Plugin manifest
skills/do-it/SKILL.md	Skill trigger and documentation
agents/*.agent.md	Agent definitions

1. Install the plugin

Copilot CLI:

copilot plugin install brian-kelley-intel/do-it-agent

VS Code:

1. Open Preferences → Open User Settings (JSON) and add the marketplace:

   "chat.plugins.marketplaces": [
       "brian-kelley-intel/do-it-agent"
   ],

2. In the Copilot chat panel, type /plugins and install do-it-agent.

2. Use it

Ask Copilot to implement something:

• "Implement a REST API for user authentication with JWT tokens"
• "Add a dark mode toggle to the settings page"
• "Refactor the database layer to use connection pooling"
• "Do it — build the feature described in issue #42"

The orchestrator will analyze, plan, implement, and grade — then open a PR.

3. Monitor progress

Check the run state at any time:

cat .do-it/runs/*/state.json

Review plan artifacts in .do-it/runs/<run-id>/plan/.

Design Principles

1. Simplest complete solution — every decision favors simplicity over cleverness
2. Wave-based implementation — features are built and tested incrementally
3. Quality-gated — grader must pass before PR is created
4. Transparent state — all inter-agent communication is visible in .do-it/
5. Fail-safe — errors are surfaced, never swallowed; max iteration caps prevent loops

Development

# Clone and install locally
git clone https://github.com/brian-kelley-intel/do-it-agent.git
cd do-it-agent
copilot plugin install .

# Verify installation
copilot plugin list

License

MIT