
[!IMPORTANT] ContextKit is no longer actively maintained. Its planning features have evolved into 📋 PlanKit — a lighter, plugin-based system that integrates directly with Claude Code's plugin marketplace. For quick plans without a full planning system, use Claude Code's built-in Plan mode (
/plan).Migration: Install PlanKit via the 🧩 Plugin Marketplace and run
/plan-kit:initto set up your project. PlanKit covers Ideas → Roadmap → Steps → Done with a cleaner architecture and automatic archiving.
<p align="center"> <strong>Quick Navigation</strong><br> <a href="#-get-started">Get started</a> • <a href="#%EF%B8%8F-how-it-works">How it works</a> • <a href="#-commands-overview">Commands</a> • <a href="#-universal-but-opinionated">Universal platform support</a> • <a href="#%EF%B8%8F-comparison-with-other-ai-systems">Comparison</a> </p>
🤔 Why ContextKit exists
The real problem: AI assistants are reactive, not proactive. You have to spoon-feed them every single step: "Now create the model", "Now add the service", "Don't forget tests", "Now wire up the UI". They wait for instructions instead of understanding the bigger picture and driving the process forward.
The frustration: You're managing the AI instead of collaborating with it. You spend your time figuring out what to ask for next, breaking down work into AI-digestible chunks, and constantly steering the conversation. It's like having a junior developer who needs step-by-step instructions for everything.
The solution: ContextKit gives AI the planning intelligence to understand your project, suggest the right technical approach, break down features systematically, and guide you through proven development phases. The AI becomes a proactive partner that knows what comes next and why.
And you stay in control: ContextKit enables AI autonomy at the right moments while keeping you in strategic control. You review and approve the spec, tech decisions, and implementation plan at key checkpoints. The AI then executes with specialized quality agents (accessibility, localization, code cleanup) working autonomously within your approved framework.
The key insight: By helping AI understand your intent clearly upfront, it can work longer and more effectively without constant corrections. You spend time on strategic decisions, not fixing misunderstood implementations.
⚡ Get started
1. Install ContextKit globally
curl -fsSL https://raw.githubusercontent.com/FlineDev/ContextKit/main/install.sh | bash2. Setup your project
Navigate to your project directory and run:
# Start Claude Code
claude
# Initialize ContextKit (auto-detects your project type)
/ctxk:proj:init3. Choose your workflow
For major features (multi-file changes, new architecture, research needed):
# Full systematic workflow with detailed planning
/ctxk:plan:1-spec # Business requirements
/ctxk:plan:2-research-tech # Technical architecture
/ctxk:plan:3-steps # Implementation breakdown
/ctxk:impl:start-working # Start developmentFor smaller tasks (bug fixes, improvements, refactoring):
# Quick workflow with interactive validation
/ctxk:plan:quick # All-in-one: understanding → plan → workContextKit creates structured feature directories, maintains context between sessions, and guides you through quality checkpoints that prevent common AI development pitfalls.
[!NOTE] Early Development: ContextKit is actively evolving with iterative improvements. While production-ready (I use it daily for all my projects), expect some rough edges as commands and agents are refined. ContextKit updates itself globally when you start new chats, and the
/ctxk:proj:migratecommand makes updating your project files easy - keep sending feedback to help iron out issues faster.
⚙️ Optimized Configuration
ContextKit configures Claude Code with optimized settings for systematic development:
- Sonnet Model Default: Sufficient quality with proper guidance while reducing 5h-limit risk during complex planning phases
- Custom Statusline: Real-time monitoring so you have full transparency about the limits:
Progress bar colors: Light Gray (<50%), Yellow (50-80%), Red (>80%) for visual context awareness
5h-Usage: 73% (1.4h left) | Chat: ▓▓▓▓▓▓░░░░ 64% (128k/200k)
[!TIP] Disable Auto-Compact for Better Control
Type
/configin Claude and set "Auto-compact" to false. Auto-compact triggers at ~85% context usage, often interrupting you when nearly finished with a task and plenty of relevant context available.Better workflow: Start fresh chats when needed and resume with
/ctxk:impl:start-working(within feature branches after planning) – ContextKit's structured approach makes this seamless. Manual/compactloses unpredictable amounts of context, while fresh starts preserve your systematic progress.
🏗️ How it works
ContextKit implements a proven 4-phase methodology that prevents the chaos of unstructured AI development:
Phase 1: Business Case (/ctxk:plan:1-spec)
Define what you're building and why, without technical details:
- User stories and acceptance criteria
- Success metrics and validation points
- Explicit uncertainty marking (no AI guessing)
- Scope boundaries (what's included, what's not)
Phase 2: Technical Architecture (/ctxk:plan:2-research-tech)
Plan how you'll build it, with constitutional compliance:
- Technology choices with rationale
- Swift/SwiftUI patterns and architectural decisions
- Accessibility, localization, and privacy considerations
- Integration with existing project patterns
Phase 3: Implementation Tasks (/ctxk:plan:3-steps)
Break down execution into trackable work:
- File-by-file implementation tasks with S001-S999 numbering
- Parallel execution markers ([P]) for concurrent development
- Dependency chains and validation checkpoints
- Test-driven development task ordering
Phase 4: Development (/ctxk:impl:start-working)
Execute with supervised autonomy and specialized quality agents (requires feature branch + completed planning phases):
- AI implements within your approved technical framework
- Specialized agents handle quality checks (accessibility, localization, modern code)
- Automatic cleanup of AI artifacts (leftover comments, temporary code)
- You maintain strategic oversight while AI handles tactical execution
Quick Workflow (/ctxk:plan:quick)
Streamlined single-file planning for smaller tasks (no feature branches required):
- Interactive validation - AI presents understanding of your task for confirmation
- Scope clarity - Clear In Scope/Out of Scope boundaries prevent scope creep
- Edge case identification - Surface potential issues upfront
- Fast execution - Single command, single file, ready to work
When to use Quick vs Full:
| Use Quick Workflow | Use Full Workflow |
|---|---|
| Bug fixes | New major features |
| Smaller improvements | Multi-component changes |
| Refactoring | Unclear requirements needing research |
| Focused changes | Complex architectural decisions |
| Experiments | Team review required |
File structure difference:
Context/Features/
├── 001-UserAuthentication/ ← Full feature (folder)
│ ├── Spec.md
│ ├── Tech.md
│ └── Steps.md
├── 002-FixLoginButton.md ← Quick task (single file)
├── 003-OfflineSync/ ← Full feature (folder)
│ ├── Spec.md
│ ├── Tech.md
│ └── Steps.md
└── 004-AddDarkMode.md ← Quick task (single file)Both workflows maintain chronological order with sequential numbering, making it easy to track development history.
📋 Commands overview
Project Management
/ctxk:proj:init- One-command project initialization with auto-detection
Systematic Feature Development
- Full planning workflow:
/ctxk:plan:1-spec→/ctxk:plan:2-research-tech→/ctxk:plan:3-steps - Quick planning workflow:
/ctxk:plan:quick(single-file, interactive validation) - Development:
/ctxk:impl:start-workingwith autonomous quality checks - Release:
/ctxk:impl:release-appfor iOS/macOS,/ctxk:impl:release-packagefor Swift packages
Backlog & Team Management
Note: /ctxk/bckl:* commands are relatively new and might get major refinements very soon
- Idea capture:
/ctxk:bckl:add-ideawith evaluation framework - Bug reporting:
/ctxk:bckl:add-bugwith impact assessment - Prioritization:
/ctxk:bckl:prioritize-ideasand/ctxk:bckl:prioritize-bugswith binary search evaluation - Completion:
/ctxk:bckl:remove-ideaand/ctxk:bckl:remove-bugfor cleaning backlog - Workspace setup:
/ctxk:proj:init-workspacefor client/team standards (to run in parent folder containing multiple projects)
Project Management
/ctxk:proj:init- Setup project context by auto-detecting type and configuring ContextKit/ctxk:proj:init-workspace- Configure team standards for different clients/companies/ctxk:proj:migrate- Upgrade to latest ContextKit versions
Systematic Feature Planning
/ctxk:plan:1-spec- Define requirements with business specifications and user stories (full workflow)/ctxk:plan:2-research-tech- Design architecture with technical planning and constitutional compliance (full workflow)/ctxk:plan:3-steps- Create numbered, trackable implementation tasks (full workflow)/ctxk:plan:quick- Quick planning with interactive validation for smaller tasks (single file)
Quality-Driven Implementation
/ctxk:impl:start-working- Begin development with context-aware guidance/ctxk:impl:commit-changes- Commit changes with smart formatting and message generation/ctxk:impl:release-app- Release to AppStore with iOS/macOS workflow automation/ctxk:impl:release-package- Publish Swift packages with release workflow
Backlog & Issue Management
/ctxk:bckl:add-idea- Add feature ideas with structured evaluation framework/ctxk:bckl:add-bug- Report bugs with impact assessment and reproduction steps/ctxk:bckl:prioritize-ideas- Prioritize ideas through strategic organization/ctxk:bckl:prioritize-bugs- Triage bugs with severity-based prioritization/ctxk:bckl:remove-idea- Remove completed or cancelled ideas from backlog/ctxk:bckl:remove-bug- Remove fixed or resolved bugs from backlog
🤖 Built-in Quality Agents
ContextKit includes specialized sub-agents that automatically ensure code quality at the right moments while preventing your context from getting overloaded:
-
build-project- Execute builds and report results without filling context with verbose output -
run-test-suite- Execute complete test suite with build validation and structured failure reporting -
run-specific-test- Execute specific test with build validation and focused failure analysis -
[NEEDS REWORK]
check-accessibility- VoiceOver labels, color contrast, keyboard navigation, inclusive design patterns -
[NEEDS REWORK]
check-localization- String Catalog validation, cultural adaptation, region formatters, pluralization -
[NEEDS REWORK]
check-error-handling- ErrorKit patterns, typed throws, user-friendly error messages -
[NEEDS REWORK]
check-modern-code- Replace outdated APIs (Date() → Date.now, TimeInterval → Duration, async/await patterns) -
[NEEDS REWORK]
check-code-debt- Remove AI artifacts, consolidate patterns, extract reusable components -
... more to be added over time with updates!
These agents activate automatically during development phases,
…