convez.ai
Skill
luv-jeri

tool-factory-skills

luv-jeri/tool-factory-skills

Ver no GitHub
InicialPluginScore · de 0 a 100

Evidence-gated Claude Code skill pipeline for shipping micro web tools: pick-next-tool, competitive analysis, PRD generator (+9 planned). IRON LAWS, fail-closed engines, adversarial passes, RED-GREEN evals.

Instalação

$npx convez-ai luv-jeri/tool-factory-skills

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
2

README

A tradução para o português deste README ainda está na fila.

tool-factory-skills

A single repository of Claude Code skills that together form an evidence-gated pipeline for shipping small, single-purpose, ad-monetized web tools — from "what should we build next?" all the way to a machine-readable build contract, with build/SEO/ship/monetize skills on the roadmap.

Structured as one plugin with many sub-skills, in the spirit of anthropics/skills and obra/superpowers.

The pipeline

Each stage is a separate skill with a strict input/output contract. The output of one stage is the only legal input of the next, so nothing is built on an unmeasured assumption.

#SkillQuestion it answersKey outputs
1pick-next-toolWhat should we build next?Verified winner + ordered roadmap (deterministic score.py, kill-gates A–D, adversarial kill pass)
2projects-competitive-analysisWhy will we win?Measured COMPETITIVE-BRIEF + source ledger + prd_seed handoff (11 measurement pillars, incl. voice-of-customer pain points and input ergonomics)
3projects-prd-generatorWhat exactly do we build?FINAL-stamped PRD.md + machine build-spec.json (fail-closed prd_lint.py + adversarial skeptic)

Planned next stages (same hardened pattern — see ROADMAP.md): design-md, task-splitter, build-from-spec, seo-content, legal-crawl, ship, qa-gate-runner, index-submit, monetize-adsense.

What "hardened" means here

Every skill in this repo follows the same discipline:

  • IRON LAWS — explicit, loophole-closed rules; violating the letter is violating the spirit.
  • Fail-closed engines — deterministic Python/JS validators (score.py, competitor_strength.py, gap_opportunity.py, prd_lint.py) that RAISE on missing inputs; "insufficient data" and "REFUSE" are legal outputs.
  • Evidence tiers — real-measured > triangulated > reasoned/UNVERIFIED; weak tiers can never back a committed decision or a v1 must-have.
  • Adversarial kill passes — a separate skeptic challenges every load-bearing claim before anything is stamped FINAL.
  • RED-GREEN evals — every skill ships evals/evals.json cases written from real, documented failures (the RED), so regressions are testable.
  • ADRs — docs/adr/ records the why behind every rule.

Install

Option A — copy or symlink individual skills (simplest)

git clone https://github.com/luv-jeri/tool-factory-skills
# per skill, into a project:
ln -s /path/to/tool-factory-skills/skills/pick-next-tool  your-project/.claude/skills/pick-next-tool
# (or cp -R instead of ln -s)

Skills then invoke by bare name: /pick-next-tool, /projects-competitive-analysis, /projects-prd-generator.

Option B — as a Claude Code plugin

The repo carries .claude-plugin/plugin.json, so it can be installed as a plugin (skills invoke namespaced, e.g. /tool-factory:pick-next-tool). See the Claude Code plugin docs for marketplace setup.

Per-skill setup

pick-next-tool and projects-competitive-analysis have optional free-API integrations (Google Ads, OpenPageRank, SerpApi, Bing) — each skill's SETUP.md documents the one-time keys. All skills run in a free, browser-driven mode with zero setup.

History

These skills started life as three standalone repos (pick-next-tool, projects-competitive-analysis, projects-prd-generator) and were merged here with their git histories preserved (git subtree). This repo is now the canonical home.

License

MIT — see LICENSE.

Ver o original no GitHub

Bom para

Relacionados