The Problem: After 6 months of using Claude Code for a production project, I hit a wall: AI assistants lose context between sessions. Every morning, I'd spend 20+ minutes explaining what we built yesterday, architectural decisions, and why certain patterns were chosen. Quality degraded over time, tests became inconsistent, and I was effectively doing manual "handoffs" to my AI pair programmer.
What I Built: Solokit is a Session-Driven Development framework that treats AI coding sessions as stateful development work with perfect context continuity. Each session starts with a comprehensive 9-section briefing: work spec, docs, stack info, file tree, git context, accumulated learnings, dependencies, milestone progress, and previous work.
How It Works:
1. sk init - Choose from 4 production templates (T3 Stack, FastAPI, Refine, Next.js) with 4 quality tiers
2. sk work-new - Create work items with specs and dependency tracking
3. /start in Claude Code - AI receives full project briefing
4. Build with AI - Complete context, no repetition
- AI-powered learning categorization with smart deduplication
- Dependency visualization with critical path analysis
- Spec-first architecture (markdown specs, not JSON metadata)
- 16 slash commands integrated with Claude Code
Why It's Different:
- Not just templates - complete workflow from init to deployment
- Not just project management - AI-native with perfect briefings
- Not just quality gates - knowledge accumulation system
- Built for solo developers achieving team-level sophistication
Tech Stack: Python 3.9+, published on PyPI (pip install solokit), MIT licensed.
I've been using it to build my own SaaS for 4 months. The biggest win: I never lose context between sessions anymore. Claude knows exactly where we left off, what patterns we're following, and why decisions were made.
The framework is production-ready (v0.1.2), extensively tested, and designed for long-running projects (50+ sessions).
Try it:
pip install solokit
sk init
sk work-new
/start
Happy to answer questions about the architecture, testing strategy, or how the briefing system works!
Hey HN! I'm the creator of Solokit.
The Problem: After 6 months of using Claude Code for a production project, I hit a wall: AI assistants lose context between sessions. Every morning, I'd spend 20+ minutes explaining what we built yesterday, architectural decisions, and why certain patterns were chosen. Quality degraded over time, tests became inconsistent, and I was effectively doing manual "handoffs" to my AI pair programmer.
What I Built: Solokit is a Session-Driven Development framework that treats AI coding sessions as stateful development work with perfect context continuity. Each session starts with a comprehensive 9-section briefing: work spec, docs, stack info, file tree, git context, accumulated learnings, dependencies, milestone progress, and previous work.
How It Works:
1. sk init - Choose from 4 production templates (T3 Stack, FastAPI, Refine, Next.js) with 4 quality tiers
2. sk work-new - Create work items with specs and dependency tracking
3. /start in Claude Code - AI receives full project briefing
4. Build with AI - Complete context, no repetition
5. /end - Automated quality gates (tests, linting, security, docs) + learning capture
6. Next session - Perfect continuity, AI knows everything
Technical Highlights:
- 2,417 tests passing (100%), 85% code coverage
- Type-safe with mypy strict mode
- 8 automated quality checkers (testing, security, linting, performance, docs)
- AI-powered learning categorization with smart deduplication
- Dependency visualization with critical path analysis
- Spec-first architecture (markdown specs, not JSON metadata)
- 16 slash commands integrated with Claude Code
Why It's Different:
- Not just templates - complete workflow from init to deployment
- Not just project management - AI-native with perfect briefings
- Not just quality gates - knowledge accumulation system
- Built for solo developers achieving team-level sophistication
Tech Stack: Python 3.9+, published on PyPI (pip install solokit), MIT licensed.
I've been using it to build my own SaaS for 4 months. The biggest win: I never lose context between sessions anymore. Claude knows exactly where we left off, what patterns we're following, and why decisions were made.
The framework is production-ready (v0.1.2), extensively tested, and designed for long-running projects (50+ sessions).
Try it:
Happy to answer questions about the architecture, testing strategy, or how the briefing system works!GitHub: https://github.com/ankushdixit/solokit
Docs: https://github.com/ankushdixit/solokit/tree/main/docs