agentsclimarketplace

Spec first development

Skill roronoazoroshao369/vibe-coding-os/skills/core/spec-first-development

Vibe Coding OS — Claude/Codex/Cursor skill framework with 139 skills, 111 commands, 95 templates, 22 tracked sources, 28/28 validation gates PASS. Quality Shield, Engineering Discipline Pack, plugin marketplace.From the repository description

Install
npx -y skills add roronoazoroshao369/vibe-coding-os --skill spec-first-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

  • 0 stars0 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

7.3 KB, ~1.6k tokens by cl100k_base, as published. Nobody here has run it

Spec-First Development

Purpose

Turn intent into a compact specification before non-trivial code changes.

When to use

Use for new features, behavior changes, public APIs, data changes, or risky refactors.

Inputs

Intent, constraints, users, existing behavior, acceptance criteria.

Workflow

  1. Create or update a spec using the template.
  2. Define goals, non-goals, and constraints.
  3. Describe expected behavior, user scenarios, and edge cases.
  4. Add acceptance criteria that can be verified.
  5. Apply what-before-how: keep technical choices out of the spec.
    • Restate the desired outcome in user-visible terms: who, what they observe, and why.
    • List acceptance criteria that are observable and testable, independent of technology.
    • Identify and quarantine any "how" content (stack, schema, libraries) into the plan, not the spec.
    • Flag premature technical lock-in and ask whether it is a real constraint or an assumption.
    • Only after the what is agreed, allow technical-context decisions in the plan phase.
  6. Review the spec for missing assumptions before planning, then clear the spec checkpoint.

Outputs

A concise spec with goals, non-goals, behavior, edge cases, and acceptance criteria. A behavior-focused spec section plus a separate list of deferred technical decisions to resolve during planning.

Scenario pattern taxonomy

User scenarios in a spec should be classified by what they test, not by who writes them. Each category demands a different kind of scrutiny:

  • Happy path — the intended, sunny-day flow. Validates that the feature works when everything goes right. Every spec must have at least one happy-path scenario that traces from trigger to expected outcome without deviations.
  • Error path — what happens when inputs are invalid, dependencies fail, or preconditions are unmet. Error paths must name the specific error condition, the observable failure signal (exception, error code, degraded UI), and the recovery or fallback behavior.
  • Edge case — boundary conditions, empty states, maximums, minimums, concurrency races, or timezone/corner inputs. Edge cases reveal assumptions the happy path hides. List concrete boundary values, not abstract descriptions.
  • Performance expectation — throughput, latency, or resource-usage bounds under stated load. Quantified with a specific measurement (e.g., "response under 200 ms at 1000 req/s") rather than vague quality claims.
  • Security boundary — authorization checks, data isolation, injection resistance, or rate-limiting behavior. Security scenarios describe what an attacker or untrusted caller must NOT be able to do, and what observable signal is produced when a boundary is crossed.

A well-formed spec covers at least the happy path and at least one error path. Omitting a category is a deliberate choice that must be recorded as a non-goal.

Scenario validation checklist

For each scenario in the spec, verify:

  • The actor, trigger, and expected outcome are concretely named.
  • The scenario belongs to a recognized category (happy, error, edge, performance, security) — or the omission is explained in non-goals.
  • Acceptance criteria for the scenario are observable and can be verified by a test or inspection.
  • The scenario does not assume a particular technical implementation (what-before-how).
  • Error-path scenarios describe both the failure signal and the recovery or fallback behavior.

Apply this checklist during the spec-review checkpoint. A scenario that fails any check should be reworked or explicitly deferred before the plan phase begins.

Failure modes

  • Spec becomes too broad.
  • Acceptance criteria are not testable.
  • Implementation starts before agreement on behavior.
  • Scenarios describe implementation details instead of observable behavior.
  • A scenario category is missing without an explicit non-goal.
  • Performance expectations are stated without measurement criteria.
  • Acceptance criteria are written in implementation terms.
  • A favored technology silently constrains the spec.
  • "What" and "how" are mixed, making the spec hard to review or change.

Verification checklist

  • Goals and non-goals are explicit.
  • Acceptance criteria are observable.
  • Open questions are listed.
  • Spec matches the user request.
  • Scenario validation checklist has been applied to every scenario.
  • Acceptance criteria are observable and tech-agnostic.
  • Technical choices are deferred to the plan.
  • Premature lock-ins are flagged as assumptions or real constraints.
  • The spec can be satisfied by more than one implementation.

Related mattpocock-inspired skill: skills/core/prd-from-context/SKILL.md for product-shaped specs from existing conversation context.

Spec-driven development layer (github/spec-kit inspiration)

This skill is the specify phase of the local spec-driven workflow. For the full lifecycle, compose with:

  • skills/core/project-constitution/SKILL.md — principles that constrain the spec.
  • skills/core/acceptance-criteria/SKILL.md — observable, verifiable criteria.
  • skills/core/plan-from-spec/SKILL.md — turn the agreed spec into a plan.
  • skills/core/task-breakdown-from-plan/SKILL.md and skills/core/dependency-aware-task-ordering/SKILL.md — decompose and order tasks.
  • skills/core/checkpoint-validation/SKILL.md — the implementation-readiness gate.
  • skills/core/brownfield-spec-enhancement/SKILL.md — spec discipline for existing systems.

Applied / Not Applied

  • Applied: the what-before-how discipline from github/spec-kit (merged from the former skills/core/spec-first-development/SKILL.md in v2.17 trim).
  • Not applied: upstream phrasing, command names, or template text. The separation is enforced locally through the spec and plan skills.

Inspiration source and mapping: references/sources/github-spec-kit.md, references/features/spec-driven-development.md. No upstream content is vendored, and the Specify CLI is not required.

Ghi chú tiếng Việt

Skill này là pha specify. Kết hợp với các skill spec-driven khác (constitution, what-before-how, acceptance-criteria, plan-from-spec, task breakdown, checkpoint) để có vòng đời đầy đủ. Học ý tưởng từ spec-kit, không copy, không bắt buộc Specify CLI.

"Cái gì trước, làm thế nào sau" (what-before-how): thống nhất hành vi người dùng và tiêu chí chấp nhận (what) trước khi chọn công nghệ (how). Đẩy mọi quyết định kỹ thuật sang pha plan. Học ý tưởng từ spec-kit, không copy chữ. Liên kết: skills/core/plan-from-spec/SKILL.md.

Choose instead

Greenfield projects where the contract must be locked before any code.

If this skill is not the right phase for your task, use one of these instead:

  • skills/core/brownfield-spec-enhancement/SKILL.md — if a spec already exists and you're extending it.
  • skills/core/writing-plans/SKILL.md — if no spec is needed (small task, single author, throwaway).
  • skills/core/brainstorming/SKILL.md — if the direction is not yet chosen.

For the full decision tree across all 6 plan-family skills, see docs/workflows/plan-skill-decision-tree.md.

What ships with it

Read from the repository

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

Keep looking

Skills are one crate of 326,750. 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.