spec-first-mvp — Claude Code skill: what it does, how to install
The spec-first-mvp skill stops development before implementation and requires a short spec, a list of throwaway probes, and a go/no-go decision point.
no license ★ 475 updated 2026-10-13
npx skills add cnfjlhj/ai-collab-playbook --skill spec-first-mvp -a claude-codeWhat spec-first-mvp does
The skill inserts a validation step between idea and code. It forces the agent to turn the initial idea into a spec seed: the problem, user context, first hypothesis, explicit non-goals, and a pass/fail signal. Then the agent proposes 1–3 throwaway probes — a static page, one prompt with example I/O, or a minimal scoring script. Each probe tests one risky assumption.
After running the probe, the agent updates the spec with evidence rather than expanding the original fantasy unchanged. Only when the result is measured and the go/no-go decision is made does the skill hand off control — to brainstorming, planning, or direct execution. This is a text workflow, no external tools required.
The skill is useful for tools, research helpers, personal knowledge systems, and product concepts whose shape is still unstable.
What it helps with
- stop implementation before the scope is stable
- separate the real problem from guesses and wishful extras
- test a risky assumption with a throwaway probe
- update the spec with evidence after an experiment
- make a go/no-go decision based on a measurable signal
Example prompts for the agent
- Stop, don't write code yet — first turn this idea into a spec seed
- Propose three throwaway probes to test whether this concept works
- Update the spec with what we learned from the probe and decide whether to continue
How to install spec-first-mvp
1Skills CLI (Vercel)
npx skills add cnfjlhj/ai-collab-playbook --skill spec-first-mvp -a claude-code2Manually: put the files in place
~/.claude/skills/spec-first-mvp/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/spec-first-mvp/
1Skills CLI (Vercel)
npx skills add cnfjlhj/ai-collab-playbook --skill spec-first-mvp -a codex2Manually: put the files in place
.agents/skills/spec-first-mvp/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add cnfjlhj/ai-collab-playbook --skill spec-first-mvp -a cursor2Manually: put the files in place
.cursor/skills/spec-first-mvp/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install cnfjlhj/ai-collab-playbook skills/full/spec-first-mvp2Manually: put the files in place
.github/skills/spec-first-mvp/SKILL.md1Skills CLI (Vercel)
npx skills add cnfjlhj/ai-collab-playbook --skill spec-first-mvp -a gemini-cli2Platform CLI
gemini skills install https://github.com/cnfjlhj/ai-collab-playbook --path skills/full/spec-first-mvpManually: put the files in place
.devin/skills/spec-first-mvp/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about spec-first-mvp
How to install spec-first-mvp in Claude Code?
Run npx skills add cnfjlhj/ai-collab-playbook --skill spec-first-mvp -a claude-code, the skill will appear in ~/.claude/skills/spec-first-mvp/SKILL.md.
What does spec-first-mvp do?
It inserts a validation step between idea and code: the agent writes a spec seed, proposes throwaway probes to test risky assumptions, and makes a go/no-go decision based on measurable evidence.
When should I use spec-first-mvp?
When the user asks to "just build it" but requirements are still drifting, or when you need to validate one hypothesis before writing full code. For comparing multiple directions, use a different skill.