Skip to content

ROADMAP.md

Latest commit

 

History

History
127 lines (98 loc) · 3.36 KB

ROADMAP.md

File metadata and controls

127 lines (98 loc) · 3.36 KB

CodeCourt - Development Roadmap

Overview

Multi-agent PR review system with auto-fix, security scanning, and debate features.

Phase 1: Foundation ✅

1.1 Project Setup

  • Initialize monorepo structure
  • Set up Python package with pyproject.toml
  • Set up TypeScript package with package.json
  • Create docs/ROADMAP.md, PROGRESS.md, devlog/

1.2 LLM Provider Abstraction

  • Create base provider interface (providers/base.py)
  • Implement OpenAI provider
  • Implement Anthropic provider
  • Implement Ollama provider
  • Add provider factory with config

1.3 Git/Diff Tools

  • Parse git diff output
  • Extract changed files and hunks
  • Read file contents with context
  • Create patch objects

1.4 Basic Code Review Agent

  • Single agent that reviews a diff
  • Returns structured feedback (file, line, severity, message)
  • Test on sample PRs

Phase 2: Multi-Agent Core ✅

2.1 Coordinator Agent

  • Orchestrates multiple specialist agents
  • Parallel execution support
  • Aggregates and deduplicates findings
  • Prioritizes issues by severity
  • Consensus-based approval decision

2.2 Security Agent

  • OWASP vulnerability detection
  • Secrets/credential scanning
  • Dependency vulnerability awareness
  • Security-specific prompts and rules

2.3 Agent Communication

  • Define message format between agents (models.py)
  • Implement handoff protocol via Coordinator
  • Add logging for debugging agent interactions

Phase 3: Advanced Features (Current)

3.1 CLI ✅

  • Create cli.py with Click
  • codecourt review <diff-file> command
  • codecourt review --stdin for piped input
  • codecourt review --repo <path> for local repos
  • Provider/model selection (--provider, --model)
  • Output formats (--format rich|json|markdown)
  • codecourt providers command
  • codecourt parse command
  • Config file support (.codecourtrc)

3.2 Debate System

  • Defender agent: argues FOR the code changes
  • Prosecutor agent: argues AGAINST / finds problems
  • Judge agent: synthesizes and makes final call
  • Structured debate rounds (2-3 exchanges)

3.3 Auto-Fix Agent

  • Generate fix suggestions as patches
  • Validate patches apply cleanly
  • Run basic syntax checks on fixes
  • Output unified diff format

3.4 Review Report Generator

  • Aggregate all agent outputs
  • Generate markdown summary
  • Include confidence scores
  • Format for PR comment

Phase 4: CLI + GitHub Action

4.1 FastAPI Server

  • REST API for triggering reviews
  • Async processing with status polling
  • Webhook support for GitHub

4.2 TypeScript CLI

  • pr-review analyze <repo> <pr-number>
  • pr-review review --local (for local diffs)
  • Config file support (.pr-reviewrc)

4.3 GitHub Action

  • action.yml with inputs
  • Trigger on PR open/update
  • Post comments via GitHub API
  • Optional auto-fix commit

Phase 5: Polish

5.1 Testing

  • Unit tests for each agent
  • Integration tests with mock LLMs
  • End-to-end test on real repo

5.2 Documentation

  • README with setup instructions
  • Architecture documentation
  • Demo GIFs/screenshots

5.3 Demo & Deploy

  • Deploy to a sample repo
  • Record demo video
  • Write blog post / case study