A Claude Code skill for building resumable AI pipelines with tuff-lil-unit.
Tell Claude what you want to build. Tuff handles the rest — crash recovery, concurrency, retry, and a queryable SQLite database of everything that happened.
| Phase | What it does | Model | Calls | Concurrency |
|---|---|---|---|---|
| Collect | Fetch 200 source URLs | none | 200 | 50 |
| Extract | Pull structured data from each page | GPT-5-nano | 200 | 10 |
| Classify | Score and categorise each result | GPT-5-mini | 200 | 5 |
| Distil | Aggregate into final report | none | 1 | 1 |
~400 LLM calls, 200 HTTP fetches, 4 phases. One /tuff invocation.
Via the unfamiliar-city marketplace:
/plugin marketplace add unfamiliar-city/marketplace
/reload-plugins
/plugin install tuff
Or add the skill directly to your project:
mkdir -p .claude/skills
git submodule add https://github.com/unfamiliar-city/skill-tuff .claude/skills/tuff
Claude Code
Opus 4.6 · Claude API
~/Projects/myproject
> /tuff build a pipeline that scrapes 200 URLs, extracts key data
with GPT-5-nano, and generates a summary report
Claude designs the phases, sets up the file structure, installs tuff-lil-unit, and writes the code. If it crashes, re-run — it resumes from where it left off.
- Pipeline design — phases, table schemas, concurrency tuning
- The full tuff API —
ctx.step,ctx.upsert,ctx.model.*,ctx.agent.* - Three execution modes — LLM APIs (Anthropic/OpenAI), Claude Code headless (play at your own risk), any async function
- Architecture patterns — file structure, fan-out, composition, progress tracking
The skill installs tuff-lil-unit (npm i tuff-lil-unit) — a TypeScript toolkit that implements the step-function pattern with SQLite persistence. See the library repo for API docs and code examples.
Alpha. See tuff-lil-unit for details.
MIT