TL;DR
PACT is a structured workflow with specialized AI agents for each phase: research before building, design before implementing, code with checklists, and validate before shipping. 10x better results than “just ask Claude.”
The Four Phases
Why Structured AI Development?
Unstructured AI Development
- Jump straight to code → miss edge cases
- No research → reinvent the wheel
- Circular fixes → same bugs return
- No testing → ship broken code
- Lost context → repeat mistakes
With PACT Framework
- Research first → informed decisions
- Architecture plans → fewer rewrites
- Checklist-driven → predictable quality
- Built-in QA → confident shipping
- Session memory → compound learning
Prepare
Research before you build
Gather documentation, understand requirements, and build context before writing a single line of code.
Outputs
- → Research documentation
- → Competitive analysis
- → Technical requirements
Key Modules
- context7-integration
- session-memory
- deep-investigation
Agent: grit-pact-preparer
Architect
Design before you implement
Create detailed architectural plans, identify edge cases, and define the implementation approach.
Outputs
- → Architecture diagrams
- → API contracts
- → Data models
- → Implementation plan
Key Modules
- subtask-planning
- verification-types
- spec-critique
Agent: grit-pact-architect
Code
Build with confidence
Implement the architecture with specialized agents, pre-implementation checklists, and recovery tracking.
Outputs
- → Production-ready code
- → Unit tests
- → Documentation updates
- → Migration scripts
Key Modules
- pre-implementation-checklist
- self-critique
- recovery-tracking
Agents: grit-pact-backend-coder, grit-pact-frontend-coder, grit-pact-database-engineer
Test
Validate before you ship
Comprehensive testing with the Test Companion, automated QA loops, and human validation.
Outputs
- → Test reports
- → Bug fixes
- → Performance metrics
- → Sign-off documentation
Key Modules
- qa-fix-loop
- playwright-decision-tree
- validation-patterns
Agent: grit-pact-test-engineer
21 Reusable Prompt Modules
PACT's power comes from its modular architecture. Each module solves a specific problem and can be combined as needed.
.claude/grit-pact/modules/ ├── deep-investigation.md # Mandatory codebase analysis ├── complexity-assessment.md # Pre-task risk analysis ├── context7-integration.md # Library doc lookup ├── subtask-planning.md # Breaking work into units ├── pre-implementation-checklist.md ├── self-critique.md # Quality gate ├── recovery-tracking.md # Prevent circular fixes ├── session-memory.md # Cross-session persistence ├── parallel-execution.md # Multi-agent coordination ├── insight-extractor.md # Knowledge capture ├── qa-fix-loop.md # Structured fix cycles ├── spec-critique.md # Pre-coding validation ├── model-selection.md # haiku/sonnet/opus routing └── validation-patterns.md # API/Browser/DB verification
Session Memory: Learning That Persists
Unlike traditional AI workflows where context is lost, PACT maintains session memory that compounds your learning over time.
.gritflow/ ├── patterns.md # Discovered patterns (reused) ├── gotchas.md # Known issues and pitfalls ├── codebase_map.json # Project structure ├── attempt_history.json # What's been tried └── session_insights/ # Per-session learnings
Start Using PACT
Run your first PACT workflow with the full guide on the dedicated PACT page.