Workflow Overview

7-Step Workflow Overview

The antidote to vibe coding. Each step is a Claude Code skill, maps to one of Covey's 8 Habits, and has an explicit handoff contract with the next step.

The 7 Steps (+ Step 0)

Legend:  [O] optional   [H] human checkpoint   [!] NEVER SKIP

   [O] 0 · /research        H5 Understand    investigate before specifying
        │
        ▼
   [H] 1 · /requirements    H2 End in Mind   define what, why, who
        │
        ▼
   [H] 2 · /design          H8 Voice         architecture (human-led)
        │
        ▼
   [O] 3 · /breakdown       H3 First Things  atomic tasks, no scope creep
        │
        ▼
   [O] 4 · /build-brief     H5 Understand    context before coding
        │
        ▼
   [!] 5 · /review-ai       H4 Win-Win       audit before commit
        │
        ▼
   [H] 6 · /deploy-guide    H1 Proactive     staging first, rollback ready
        │
        ▼
   [O] 7 · /monitor-setup   H7 Sharpen Saw   observe after deploy

Step	Skill	Habit	Output	Human checkpoint?
0	/research	H5 Understand First	Research brief	Optional
1	/requirements	H2 Begin with End	PRD summary	Yes
2	/design	H8 Find Your Voice	Architecture decisions (ADR)	Required
3	/breakdown	H3 First Things First	Atomic task list	Optional
4	/build-brief	H5 Understand First	Implementation brief per task	Optional
5	/review-ai	H4 Win-Win	Code review findings	Required
6	/deploy-guide	H1 Be Proactive	Deploy plan + rollback	Required
7	/monitor-setup	H7 Sharpen the Saw	Observability config	Optional

Handoff Contracts

Each skill declares prev-skill and next-skill in its frontmatter and explicitly documents:

• Expects from predecessor: what input the skill needs
• Produces for successor: what output the next step will consume

This prevents context loss between steps and makes the workflow composable.

Note

The handoff DAG is machine-verified by tests/test-skill-graph.sh — no cycles, no dangling references, symmetric edges.

When to Skip

Change type	Steps to run
Single-line bug fix (obvious root cause)	5
Formatting / linting	5
Dependency version bump	5, 6
New feature (small, <3 files)	1, 3, 4, 5, 6
New feature (medium)	1, 2, 3, 4, 5, 6, 7
New feature (complex / unclear domain)	0, 1, 2, 3, 4, 5, 6, 7
Refactor (no behavior change)	2, 3, 4, 5
Architecture change	0, 1, 2, 3, 4, 5, 6, 7

Warning

Never skip /review-ai (Step 5). Every other step has legitimate skip cases; review does not.

Beyond the Workflow

These skills complement the pipeline and can run at any point:

Assessment:

• /cross-verify — 17-question 8-Habit checklist
• /whole-person-check — Body/Mind/Heart/Spirit balance
• /security-check — OWASP Top 10 focused review
• /reflect — post-task retrospective with lesson persistence
• /workflow — interactive guided walkthrough

Meta & Onboarding:

• /using-8-habits — onboarding decision tree (no-arg) or smart-routing mode when invoked with intent (e.g. /using-8-habits "I need to verify what we built last week" returns ≤3 ranked skills with reasoning)
• /calibrate — maturity self-assessment, verbosity adaptation

Compliance & Audit:

• /eu-ai-act-check — EU AI Act 9-obligation checklist
• /ai-dev-log — AI development log from git history

See also

• Skills Catalog — all 17 skills with when-to-use guidance
• Habits Reference
• Vibe Coding vs Structured