Files
fete/WORKFLOW.md
nitrix 1afdd0a462
Some checks failed
CI / backend-test (push) Successful in 59s
CI / frontend-test (push) Failing after 16s
CI / frontend-e2e (push) Successful in 51s
CI / build-and-publish (push) Has been skipped
Add spec-kit workflow guide
Human-readable reference for the clarify → plan → tasks → analyze → implement
loop and the file structure it produces.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-03-06 20:22:39 +01:00

2.6 KiB

Spec-Kit Workflow

How to take a feature from spec to working code.

The Loop

clarify → plan → tasks → analyze → implement

Every step produces files in specs/NNN-feature-name/. Review after each step before moving on.

Steps

1. Clarify the spec (optional)

/speckit.clarify

Asks up to 5 targeted questions about gaps or ambiguities in the spec. Answers get folded back into spec.md. Skip this if the spec is already solid.

2. Create the implementation plan

/speckit.plan

Reads spec.md, checks it against the constitution, and produces:

File Content
plan.md Technical design, architecture decisions, contracts
research.md Background research (if needed)
data-model.md Entity definitions (if new entities are involved)
contracts/ API contracts (if applicable)

Review this carefully. The plan shapes everything downstream.

3. Generate the task list

/speckit.tasks

Reads the plan and generates tasks.md — an ordered, dependency-aware task list. Tasks marked [P] can run in parallel.

4. Check consistency (optional)

/speckit.analyze

Cross-checks spec, plan, and tasks for contradictions, gaps, or drift. Non-destructive — only reports, doesn't change files.

5. Implement

/speckit.implement

Executes the tasks from tasks.md phase by phase. Follows TDD: writes tests first, then implementation. Stops at checkpoints so you can verify.

File structure

Each feature lives in its own directory:

specs/
  007-view-event/
    spec.md              # What and why (from /speckit.specify or migration)
    plan.md              # How (from /speckit.plan)
    research.md          # Background research (from /speckit.plan)
    data-model.md        # Entity definitions (from /speckit.plan)
    contracts/           # API contracts (from /speckit.plan)
    tasks.md             # Ordered task list (from /speckit.tasks)

Starting a brand new feature

If the feature doesn't have a spec yet:

/speckit.specify

Describe what you want in plain language. This creates the spec directory and spec.md from the template. Then continue with the loop above.

Tips

  • Don't skip the review. Each step builds on the previous one. Garbage in, garbage out.
  • The spec is the source of truth. If something in the plan contradicts the spec, fix the spec first.
  • You can re-run steps. Changed the spec after planning? Run /speckit.plan again.
  • Constitution governs everything. Principles in .specify/memory/constitution.md override ad-hoc decisions.