tool-factory-skills
luv-jeri/tool-factory-skills
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-skills1 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
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.
| # | Skill | Question it answers | Key outputs |
|---|---|---|---|
| 1 | pick-next-tool | What should we build next? | Verified winner + ordered roadmap (deterministic score.py, kill-gates A–D, adversarial kill pass) |
| 2 | projects-competitive-analysis | Why will we win? | Measured COMPETITIVE-BRIEF + source ledger + prd_seed handoff (11 measurement pillars, incl. voice-of-customer pain points and input ergonomics) |
| 3 | projects-prd-generator | What 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.jsoncases 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.