Spec-first kickoff
Turns a vague app idea into a one-page spec and a step-by-step build plan before any code is written, asking only the questions that change the design. Use when the user describes a new project, feature or app idea and wants to start building.
spec-first-kickoffInstall
Run one command — it saves SKILL.md into your agent's skills folder. Restart the agent and the skill is picked up automatically.
mkdir -p ~/.claude/skills/spec-first-kickoff && curl -fsSL https://anymodel.org/en/skills/vibe-coding/spec-first-kickoff/SKILL.md -o ~/.claude/skills/spec-first-kickoff/SKILL.md
mkdir -p ~/.agents/skills/spec-first-kickoff && curl -fsSL https://anymodel.org/en/skills/vibe-coding/spec-first-kickoff/SKILL.md -o ~/.agents/skills/spec-first-kickoff/SKILL.md
New-Item -ItemType Directory -Force "$HOME\.claude\skills\spec-first-kickoff" | Out-Null; Invoke-WebRequest https://anymodel.org/en/skills/vibe-coding/spec-first-kickoff/SKILL.md -UseBasicParsing -OutFile "$HOME\.claude\skills\spec-first-kickoff\SKILL.md"
New-Item -ItemType Directory -Force "$HOME\.agents\skills\spec-first-kickoff" | Out-Null; Invoke-WebRequest https://anymodel.org/en/skills/vibe-coding/spec-first-kickoff/SKILL.md -UseBasicParsing -OutFile "$HOME\.agents\skills\spec-first-kickoff\SKILL.md"
Any other agent: save the file into its skills folder, or give it the link to the file: https://anymodel.org/en/skills/vibe-coding/spec-first-kickoff/SKILL.md
SKILL.md
Spec-First Kickoff
Goal: replace "build me X" with a short, agreed spec and a plan. Writing code against a vague idea produces rework. Do not create project files or install anything until the spec is confirmed.
Workflow
- Restate the idea in one sentence. If the working directory already has code, skim the README, manifest files and top-level layout first, and match the existing stack instead of proposing a new one.
- Ask questions, but only the ones whose answers change the design. Maximum 5, in a single message, each with a suggested default so the user can reply "defaults are fine". Good questions:
- Who uses it, and is it one user or many (accounts, permissions)?
- Where does data live and must it survive restarts?
- Web, mobile, CLI or API only?
- What must it integrate with (payments, email, an existing API)?
- Where will it run (local only, a VPS, a serverless platform)? Do not ask about colors, naming, or anything you can decide and change later.
- Write the spec (format below). Keep it to one page. Mark every guess as an assumption.
- Write the plan: 4 to 10 steps, each one small enough to finish and verify on its own (see the output format). Step 1 is always the thinnest end-to-end slice that runs.
- Show the spec and plan to the user and wait for approval or edits. Save them as
SPEC.mdin the repo only if the user agrees.
Spec format
# <Project name>
Goal: one sentence, the outcome for the user.
Users: who, and how many at first.
Must-haves (v1)
- ...
Non-goals (not in v1)
- ...
Data
- entities and the fields that matter, where stored
Screens / endpoints
- screen or `METHOD /path` : what it does
Acceptance checks
- observable checks, e.g. "submitting the form with an empty name shows an error"
Assumptions and open questions
- ...
Plan format
1. Skeleton that runs (hello world, one route or screen) check: <command or URL>
2. Core data model + one write and one read path check: ...
3. ...
Every step names its own check: a command, a test, or something visible in the running app.
Rules for a good spec
- Must-haves fit on one screen. If there are more than about 7, split into v1 and later.
- Non-goals are mandatory. They are what stops scope from growing during the build.
- Acceptance checks are concrete and testable, not "works well" or "looks good".
- Prefer boring, common technology the user already has installed. Name the exact versions of the runtime only if they matter.
- Flag the riskiest unknown (an unfamiliar API, a data source, auth) and put a small spike for it early in the plan.
Pitfalls
- Do not interrogate. Five targeted questions with defaults beat twenty open ones.
- Do not start coding "just the skeleton" before approval; the stack choice is part of the spec.
- Do not pad the spec with generic sections (security, scalability) that the user did not need for v1. Mention them only if the idea clearly requires them, such as handling payments or personal data.
- Do not silently expand scope while writing the plan. New ideas go under non-goals or "later".
- If the user says "just start", write a three-line spec with your assumptions, state them, and begin with step 1.
Report back
End with: the spec, the numbered plan, the list of assumptions you made, and the single question you most need answered before step 1.
Source: AnyModel skills catalog · License: CC BY 4.0
Connect through AnyModel API
This skill runs on whatever model your agent uses. Point Claude Code, Codex or Cursor at AnyModel to run it on Claude, GPT, Gemini and more with one key — pay per token, no subscription.
AnyModel