duthaho
5.4 KiB
title, description
| title | description |
|---|---|
| Skills Reference | 16 skills in Claude Kit organized around the 5-phase verification-first workflow. |
Skills Reference
Claude Kit is organized around a 5-phase verification-first workflow: Investigate → Design → Implement → Verify → Ship. All 14 spine skills (plus 2 setup skills) are user-invocable as /claudekit:<name>.
Every skill has 8 required sections: Frontmatter, Overview, When to Use, Process, Rationalizations table, Evidence Requirements, Red Flags, References. The Rationalizations pattern documents the excuses an engineer makes to skip a step (verbatim) with rebuttals. The Evidence Requirements name what artifact each checkpoint must produce.
How Skills Work
Skills have trigger descriptions with keywords. When your conversation matches, the skill loads automatically:
"why is this broken?" → investigate-root-cause
"how does X work?" → map-codebase
"plan this feature" → shape-spec, write-plan
"review the plan" → plan-review (dispatches architect + experience reviewer)
"is it done?" → verification-gate
"open a PR" → code-review-loop
"cut a release" → release-and-changelog
You can also invoke any skill directly by typing /claudekit:<name>.
🔍 Investigate
Surface every fact about the system before forming a theory. Every claim has a <file:line> citation; no memory-based assertions.
| Skill | Description | Triggers On |
|---|---|---|
| investigate-root-cause | 4-phase: gather → hypothesize → test → prove. Mandatory before any fix. | "bug", "error", "broken", "why does this", stack traces |
| map-codebase | Methodical evidence-cited exploration. Produces a written map a teammate can read in 3 minutes. | "how does X work", "trace", "find where", "scope of change" |
| audit-dependencies | Dependency archaeology — what's actually used vs declared, with import-graph and reachability checks for CVEs. | "deps", "audit", "CVE", "stale package", "do we use" |
🎨 Design
Convert a vague request into a written spec, then a numbered plan, then survive review before implementation begins.
| Skill | Description | Triggers On |
|---|---|---|
| shape-spec | One-to-three-page spec with goals, non-goals, constraints, falsifiable acceptance criteria, open questions. Engineering-flavored. | "spec", "what should we build", "design this", "let's add" |
| write-plan | Numbered task list with file paths, exact test commands, dependency annotations, acceptance per task, Risks section. | "plan", "break down", "task list", "implementation order" |
| plan-review | Orchestrator: dispatches 2 reviewers in parallel, consolidates into one fix gate, applies user-selected fixes. | "review the plan", "is the plan ready", "plan-review" |
| plan-review-architecture | Scores 5 sub-dimensions 0-10 (data flow, failure modes, edge cases, test matrix, rollback). | "architecture review", "data flow", "failure modes", "rollback" |
| plan-review-experience | Scores 5 sub-dimensions 0-10 (info hierarchy, state coverage, accessibility, DX ergonomics, AI-slop avoidance). | "UX review", "DX review", "API ergonomics", "states", "accessibility" |
🔨 Implement
Ship code with red-green-refactor discipline; vertical slices behind feature flags; refactor with evidence.
| Skill | Description | Triggers On |
|---|---|---|
| test-first | Red-green-refactor with strict evidence requirements. | "implement", "fix bug", "TDD", "write the test first" |
| incremental-shipping | Vertical slices behind feature flags plus refactor-with-evidence (test/perf deltas required). | "feature flag", "incremental", "vertical slice", "rollout" |
✅ Verify
Mandatory pre-completion gate. No "tests pass — trust me." Active debugging keeps a paper trail.
| Skill | Description | Triggers On |
|---|---|---|
| verification-gate | 6-step pre-completion gate: claim → tests → negative path → non-IDE check → cross-check → sign. | "done", "complete", "ready to merge", "tests pass" |
| evidence-driven-debugging | Active-debugging companion to investigate-root-cause: instrument, capture, verdict, clean up. | "debug", "instrument", "log", "trace", "what's happening at runtime" |
🚀 Ship
Reviewable PRs with verification evidence pasted; atomic releases with diff-built changelogs.
| Skill | Description | Triggers On |
|---|---|---|
| code-review-loop | End-to-end review etiquette: requesting and receiving feedback. Dispatches code-reviewer and (on sensitive paths) security-auditor. | "code review", "PR review", "request review", "address comments" |
| release-and-changelog | SemVer hygiene plus diff-built changelogs plus atomic release commits plus post-release smoke check. | "release", "version bump", "changelog", "tag", "publish" |
⚙️ Setup (off-spine)
Used once for project bootstrap, plus session-level mode switching.
| Skill | Description | Triggers On |
|---|---|---|
| init | Interactive setup wizard — scaffolds rules, hooks, and MCP configs into your project | /claudekit:init |
To switch session behavior (Brainstorm, Implementation, Review, etc.), use Claude Code's native output styles instead of a skill — switch via /config.