agentsclimarketplace

Spec driven development

Skill manastalukdar/ai-devstudio/skills/spec-driven-development

Gated 4-phase workflow (Specify → Plan → Tasks → Implement) where nothing advances without explicit human review — surfaces assumptions before any code is written.From its SKILL.md

Install
npx -y skills add manastalukdar/ai-devstudio --skill spec-driven-development

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.3 KB, 710 tokens by cl100k_base, as published. Nobody here has run it

Spec-Driven Development

I'll walk you through four locked phases: Specify, Plan, Tasks, Implement. Nothing moves forward without your explicit "proceed" at each gate. All assumptions surface in Phase 1 before any code is written.

Arguments: $ARGUMENTS — feature, bug fix, or initiative to drive through the workflow

Token Optimization

Expected range: 500–1,500 tokens per phase; cache the spec artifact and reuse across phases

Early exit: If $ARGUMENTS already contains a written spec document, skip Phase 1 and open Phase 2

Patterns used: Git diff scope (only changed files in Phase 4), progressive disclosure (one phase at a time)

Phase 1 — Specify

Produce a spec document covering:

  • Goal: one sentence — what success looks like
  • Scope: what is in and explicitly what is out
  • Assumptions: everything assumed true that could be wrong; flag any high-risk assumption
  • Acceptance criteria: numbered, each independently verifiable
  • Open questions: items that need answers before implementation begins

Save to docs/specs/<slug>.md if a docs/ directory exists, otherwise output inline.

Gate: "Proceed to Plan?" — do not continue until the user confirms or amends the spec.

Phase 2 — Plan

With the confirmed spec, produce an architecture plan:

  • Approach: chosen design and why (one paragraph)
  • Alternatives considered: 2–3 alternatives and why they were ruled out
  • Files affected: list every file that will change, be created, or be deleted
  • Dependencies: external packages, APIs, or services required
  • Risks: top 3 risks and mitigations

Gate: "Proceed to Tasks?" — do not continue until the user confirms or amends the plan.

Phase 3 — Tasks

Break the confirmed plan into a dependency-ordered task list:

  • Each task must be independently completable and testable
  • Annotate tasks with [~30m], [~2h], etc. effort estimates
  • Mark blocking dependencies between tasks explicitly
  • Each task ends with a clear "done when" criterion

Save to tasks/<slug>-tasks.md if a tasks/ directory exists, otherwise output inline.

Gate: "Proceed to Implement?" — do not continue until the user confirms or amends the task list.

Phase 4 — Implement

Execute tasks in dependency order. For each task:

  1. State which task is starting
  2. Implement it
  3. Run the applicable test or verification command
  4. Report "Task N complete — [done-when criterion met]" before moving to the next task

Follow /incremental-implementation discipline: no more than ~100 lines of new code before verifying.

Edge Cases

  • User skips a gate ("just do it all"): Acknowledge, warn that skipping gates increases rework risk, then proceed if they confirm
  • Spec reveals the wrong problem: Surface this in Phase 1; do not silently pivot
  • Phase 4 reveals a spec gap: Pause, add the gap to the spec as an amendment, confirm with the user, then resume
  • No docs/ or tasks/ directory: Output artifacts inline; note they were not persisted

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 326,834. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.