What is it about
Hands-on workshop to treat AI as a system that compounds—not a chat that resets every session. Build CLAUDE.md, domain skills and real verification on your codebase with the Skill-Driven Development (SDD) methodology.
Skill-Driven Development — naive AI use resets to zero every session. SDD makes team knowledge compound: skills + CLAUDE.md + verification.
⚠️ The problem
- Every chat starts from zero: you re-explain conventions, domain and past mistakes
- AI "decides alone" and erodes team expertise
- Speed without a safety net → debt, subtle bugs and expensive hallucinations
💡 The solution: SDD
Treat AI as a system that compounds. You are the conductor; AI is the orchestra.
- Intelligent context — CLAUDE.md + skills = codebase DNA
- Strategic delegation — Haiku for repetitive work; Sonnet/Opus for design
- Trust, but verify — round-trips, TDD, QA checklists
- Directed synthesis — AI proposes; you decide
- Process discipline — PR-only, tests before commit
- Continuous learning — bug → LEARNINGS.md → updated skill
🧩 Syllabus
- SDD 101 — Why prompting is not enough: naive sessions start from zero; SDD turns team knowledge into persistent institutional memory
- The 6 pillars: intelligent context, strategic delegation, trust-but-verify, directed synthesis, process discipline and continuous learning
- CLAUDE.md: the codebase DNA (overview, conventions, constraints, test strategy)
- Domain skills vs methodology: Five Fingerprints to discover your first business skills
- Strategic delegation: Haiku for repetitive work; Sonnet/Opus for design
- Template pattern: Sonnet designs the first example; Haiku replicates the rest
- Verification: round-trips, TDD, QA checklists and catching hallucinations
- KCP (Knowledge Context Protocol): knowledge.yaml to orient the agent in few tool calls
- Daily SDD workflow on a real backlog task
- Go Wild Challenge + 30-day skill backlog challenge
🛠️ Workshop activities
- Baseline: analyze architecture, tests, risks and dependencies of your codebase
- Write CLAUDE.md: overview, conventions, constraints and test strategy (with peer review)
- First 3 skills: architecture + domain + QA; acid test in a fresh session
- Template pattern: Sonnet designs; Haiku executes repetitive work
- Test expansion with Haiku: coverage gaps, edge cases, round-trip tests
- QA review skill + LEARNINGS.md: quality checklist and lesson capture
- Real workflow: one backlog task done with the SDD method
- Go Wild Challenge (~90 min): build something that seemed impossible that morning
🎯 Day deliverables
- Verified CLAUDE.md (changes AI behavior)
- 3–5 domain skills that pass the acid test
- One real task completed with the SDD workflow
- LEARNINGS.md + prioritized 30-day skill backlog
📦 Requirements
- Professional software development experience (any stack)
- Laptop with Claude Code / Cursor (or similar) and access to your own or a practice codebase
- Willingness to leave with real infrastructure — not just prompting tips