Appearance
Getting Started
SpecArk is a plugin bundle for Structured Prompt-Driven Development that works with both Codex and Claude Code. This page is the fastest path to a first successful run.
Before you start
You are in the right place if you want:
- a repeatable path from idea or requirement to implementation
- file-backed artifacts between phases
- an explicit skill surface instead of vague prompting
- a clear handoff from one phase to the next
If you only need installation details, jump to Codex install or Claude install.
Step 1: Install
bash
codex plugin marketplace add maxgda/spec-ark
codex plugin marketplace upgradebash
claude plugin marketplace add maxgda/spec-ark
claude plugin install specarkSee Claude Code Installation for full details and verification steps.
After install
- The
spdd-*skills are available in your session. - You are ready to give the workflow a real input artifact.
Step 2: Pick the right first command
Most first-time users should start with the orchestrator:
text
Use the spdd-orchestrator skill on @idea-of-the-enhancement.md in semi-auto mode.text
/specark:spdd-orchestrator @idea-of-the-enhancement.md semi-autoUse a manual phase only when you already have the matching artifact:
| Your artifact | Start phase |
|---|---|
| Rough idea missing problem, user, outcome, or scope context | spdd-discovery |
| Broad idea or noisy PRD with enough context to slice | spdd-plan → spdd-story |
| Broad requirement | spdd-story |
| Focused story or requirement | spdd-analysis |
Analysis artifact in spdd/analysis/ | spdd-reasons-canvas |
Prompt artifact in spdd/prompt/ | spdd-generate |
Good signal
- You are not forcing a broad idea into
spdd-generate. - Your command matches the artifact you actually have.
- The workflow starts from the smallest valid phase instead of repeating context.
Step 3: Watch the artifact handoff
SpecArk works best when each phase creates a file that becomes the next phase's input.
idea.md
└─ spdd/discovery/DISCOVERY.md ← spdd-discovery (optional, manual)
└─ requirements/STORY-001.md ← spdd-story
└─ spdd/analysis/ANALYSIS.md ← spdd-analysis
└─ spdd/prompt/PROMPT.md ← spdd-reasons-canvas
└─ implementation ← spdd-generate
└─ spdd/tests/ ← spdd-api-test (optional)Good signal
- Each step leaves behind a repository file.
- The next command points at that file instead of pasting the same context again.
- You can inspect the artifact before moving forward.
Step 4: Know when to choose orchestrator vs manual flow
Choose the orchestrator when:
- the requirement is broad
- you want review gates between phases
- you are still learning the workflow
Choose manual phase entry when:
- you already have the correct artifact for the next phase
- you want a narrow, single-purpose request
- you are resuming work after a previous phase already completed
Step 5: Continue into your first feature
Once installation and the first artifact flow make sense, use the full end-to-end walkthrough:
That tutorial covers:
- install through first implementation
- expected output after each step
- lightweight git guidance on branches, commits, and worktrees
- how to hand the feature off for release-ready review
Common mistakes
Avoid these
- Starting with
spdd-generatebefore a prompt exists inspdd/prompt/. - Re-pasting the full requirement into every phase instead of passing the produced artifact.
- Treating README, getting started, and workflow reference pages as interchangeable.
- Mixing maintainer-only local plugin work with the first-time adopter path.