Production-grade Claude Code skills for builders who ship.
Workflow discipline meets product thinking — a curated, opinionated skill pack that bridges PM rigor with AI engineering.
The Claude Code skill ecosystem has 15,000+ repos. Most are either sprawling catalogs (300+ skills, pick your own adventure) or rigid methodologies (do it my way or not at all).
Neither helps you ship better products faster.
Skillcraft is 14 skills in 4 tiers that work as a cohesive system. Each skill enforces a specific discipline. Together, they form a workflow that moves from planning through execution to delivery — the same loop a good PM/eng hybrid runs daily, automated through Claude Code.
This is not a framework. There's no lock-in. Install one skill or all fourteen. They compose, they don't depend.
Plan before you build. Verify before you ship.
| Skill | What it enforces |
|---|---|
plan-first |
Requires a written plan before any code generation. Kills the "just start coding" impulse. |
ship-cycle |
Full development workflow: branch, implement, test, review, merge. No skipping steps. |
verify-build |
Post-implementation verification checklist. Catches what tests miss. |
clean-commit |
Enforces atomic commits, conventional messages, and PR discipline. |
task-resume |
Session continuity — picks up exactly where you left off across Claude Code sessions. |
Catch problems before they become incidents.
| Skill | What it enforces |
|---|---|
output-guard |
Validates generated code against correctness, security, and style rules before accepting it. |
simplify-code |
Reviews code for unnecessary complexity. Flags over-engineering, premature abstractions, dead code. |
debug-circuit |
Circuit breaker for debug loops. Forces a step back after 3 failed attempts instead of brute-forcing. |
craft-principles |
Enforces core engineering principles: readability, simplicity, testability, minimal dependencies. |
This is the differentiator. Product thinking as executable skills.
| Skill | What it enforces |
|---|---|
pm-docs |
Generates PRDs, one-pagers, decision logs, and launch checklists from conversation context. |
sprint-flow |
Agile execution discipline: story breakdown, acceptance criteria, sprint planning, retro prompts. |
meeting-driver |
Structures meetings with agendas, time-boxes, action items, and follow-up tracking. |
Spend context like money. Keep memory lean.
| Skill | What it enforces |
|---|---|
context-budget |
Token discipline — scoped searches, targeted reads, summarized output, and subagent delegation for heavy research. |
memory-curator |
Disciplined memory curation — what to save, how to dedupe, and when to prune stale entries. |
Install all skills:
curl -sSL https://raw.githubusercontent.com/ProjectWaja/claude-skillcraft/main/scripts/install.sh | bashInstall a single skill:
# Copy any skill folder into your Claude Code skills directory
mkdir -p ~/.claude/skills
cp -r skills/plan-first ~/.claude/skills/Verify installation:
/skills
You: "Build me a user dashboard"
Claude: *immediately generates 400 lines of code*
You: *spends 2 hours debugging, realizes it doesn't match requirements*
You: "Build me a user dashboard"
Claude (plan-first): "Before I write code, let me draft a plan:
1. Requirements: What data should the dashboard show?
2. Components: Auth check, data fetching, layout, charts
3. Acceptance criteria: What does 'done' look like?
Let's align before I write a line."
You: *approves plan, gets working code on first pass*
- AI engineers who want workflow discipline, not just code generation
- PM-turned-engineers who think in systems, not just syntax
- Solo founders who need to be both the PM and the engineer
- Teams adopting Claude Code who want guardrails without bureaucracy
- Composable, not coupled. Each skill works alone. No hidden dependencies.
- Opinionated, not rigid. Strong defaults, easy overrides.
- Product-grade, not demo-grade. Every skill tested in real shipping workflows.
- Open standard. All skills use the SKILL.md format — works on Claude Code, Codex CLI, Gemini CLI, and 30+ other tools.
Built on the Agent Skills open standard. Tested on:
- Claude Code (primary)
- Codex CLI
- Gemini CLI
See ROADMAP.md for the versioned, prioritized roadmap.
Next up:
- Cross-domain skill showcase (construction estimation, supply chain)
- Eval/test harness for skill quality scoring
npx claude-skillcraft installCLI
See CONTRIBUTING.md. PRs welcome — especially for:
- New skills that fit the PM-eng bridge thesis
- Cross-platform compatibility testing
- Security audits of existing skills
MIT. See LICENSE.
Built by ProjectWaja — PM thinking, engineer execution.