Peninsula School District's comprehensive Claude Code plugin system for AI-assisted software development.
One unified plugin combining battle-tested development workflows with self-improving meta-learning.
Version: 1.12.1 Status: ✅ Production-Ready Workflows + 🧪 Experimental Meta-Learning
# Install the marketplace
/plugin marketplace add psd401/psd-claude-coding-system
# Install the unified plugin (one command!)
/plugin install psd-claude-coding-system
# Start using immediately
/work 347 # Implement an issue
/test # Run tests
/meta_health # Check system status (after 2-4 weeks)Development Workflows (9 commands):
/work- Implement solutions + automatic security review on PR creation/architect- System architecture via architect-specialist agent/test- Comprehensive testing with coverage validation/review_pr- Handle PR feedback systematically/security_audit- Manual security audit wrapper (auto-runs in /work)/issue- AI-validated GitHub issues with latest docs + optional architecture/product-manager- Product specs → validated epic + auto-created sub-issues/compound_concepts- Find automation opportunities (manual)/clean_branch- Post-merge cleanup + automatic compound learning extraction
Meta-Learning Commands (9 commands):
/meta_health- Check system status & metrics/meta_analyze- Find patterns in your workflow/meta_learn- Get improvement suggestions/meta_implement- Apply improvements safely (with dry-run)/meta_improve- Full weekly improvement pipeline/meta_document- Auto-update documentation from code/meta_predict- Forecast future issues/meta_experiment- A/B test ideas safely/meta_evolve- Improve AI agents via genetic algorithms
Workflow Specialists (12 agents):
- NEW: architect-specialist, security-analyst-specialist
- Existing: backend-specialist, frontend-specialist, test-specialist, performance-optimizer, database-specialist, documentation-writer, llm-specialist, plan-validator, gpt-5
Meta-Learning Specialists (5 agents): meta-orchestrator, code-cleanup-specialist, pr-review-responder, document-validator, breaking-change-validator
Latest Models: All agents use claude-sonnet-4-5 and claude-opus-4-1 (Oct 2025) with extended-thinking enabled
- Getting Started (Coming soon)
- Workflow Plugin Guide (Coming soon)
- Meta-Learning System Guide ✅ - Command reference, schedules, workflows
- Architecture (Coming soon)
For detailed planning docs, see:
- Desktop:
~/Desktop/PSD-Meta-Learning-System-Plan.md(Full 6-phase plan) - Desktop:
~/Desktop/PSD-Plugin-Migration-Guide.md(Installation guide)
/plugin marketplace add psd401/psd-claude-coding-system
/plugin install psd-claude-coding-system/plugin marketplace add ~/non-ic-code/psd-claude-coding-system
/plugin install psd-claude-coding-system# Work on an issue
/work 347
# Run tests
/test
# Review PR
/review_pr 123
# Clean up after merge
/clean_branch# Let it observe for a week, then:
/meta_analyze
# See improvement suggestions
/meta_learn
# Check system health
/meta_health- ✅ 9 battle-tested development commands
- ✅ 10 workflow specialist agents
- ✅ Compound engineering principles built-in
- ✅ Latest Claude models (Sonnet 4.5, Opus 4.1)
- ✅ Extended thinking enabled for deep analysis
- 🧪 Automatic telemetry via hooks (NEW in v1.1.0)
- 🧪 Pattern detection and analysis
- 🧪 Auto-improvement suggestions with confidence scores
- 🧪 Agent evolution via genetic algorithms
- 🧪 Predictive issue detection (70%+ accuracy)
- 🧪 Living documentation generation
- 🧪 A/B testing framework for safe experimentation
- 🚀 Opus 4.5 upgrade - Architecture/planning commands use new model (80.9% SWE-bench, 66% cost savings)
- 🚀 Aggressive parallelism -
/workalways dispatches 2-3 agents simultaneously (Every's pattern) - 🚀 Pre-implementation security - Security review happens before coding, not after PR
- 🚀 Skills layer - 5 reusable workflow components for git, testing, security, telemetry, parallelism
- 🚀 Parallel context gathering -
/architectfetches context concurrently (4 parallel operations) - 🚀 Parallel PR feedback -
/review_prcategorizes and dispatches agents by feedback type - 🚀 Enhanced telemetry - Tracks parallel execution patterns, duration, agent combinations
Version: 1.8.0 Status: ✅ Production-Ready Workflows + 🧪 Experimental Meta-Learning Stability: Workflow commands battle-tested, meta-learning in active development
Aggressive Parallelism (inspired by Every's compounding-engineering plugin):
- Speed > Cost: Always dispatch 2-3 agents in parallel
- Cost: ~$1.50/issue (+$1 parallel agents, +$0.50 opus-4-5)
- Value: $23-33/issue (15-20 min rework reduction @ $100/hr)
- ROI: 15x-22x return on investment
Skills Layer - DRY principle for workflows:
- Extract common patterns (git, testing, security, telemetry, parallelism)
- Single source of truth reduces maintenance
- Reusable across commands
- Easier to enhance and test
Pre-Implementation Security:
- Security-analyst runs BEFORE coding (Phase 2.5 in
/work) - Prevents issues rather than finding them in PR review
- Reduces post-PR security findings from ~2-3 to <1 per PR
Telemetry Cleanup:
- Removed obsolete manual telemetry code from all commands
- Hook-based telemetry (v1.1.0+) handles everything automatically
We welcome contributions! Please:
- Fork the repository
- Create a feature branch
- Make your changes
- Test locally
- Submit a pull request
- Author: Kris Hagel (hagelk@psd401.net)
- Organization: Peninsula School District
- Repository: psd401/psd-claude-coding-system
MIT License - see LICENSE for details
Built with Claude Code by Anthropic, incorporating compound engineering principles for systematic improvement.
Peninsula School District - Innovating education through technology