Cookbook
Ten worked patterns — light to hard. Each one is a real, runnable command. The CLI has them built in:
tabstack recipes # browse all ten
tabstack recipes 6 # one recipe, copy-paste ready
tabstack recipes 6 --json | jq .command # agents use this form The menu
| # | Recipe | Heat | Verbs |
|---|---|---|---|
| 1 | Hacker News, but typed | 🟢 | extract |
| 2 | Docs → context | 🟢 | extract |
| 3 | The price tag | 🟢 | extract |
| 4 | Changelog roulette | 🟢 | generate |
| 5 | The star drift detector | 🟡 | extract |
| 6 | The citation machine | 🟡 | research |
| 7 | CI for facts | 🟡 | generate |
| 8 | The geo sleuth | 🟡 | extract |
| 9 | Forms with a conscience | 🔴 | automate + input |
| 10 | The post that fact-checks itself | 🔴 | all of them |
Heat guide
🟢 Light — one command, no dependencies, copy-paste and run. Good starting points.
🟡 Medium — shell loops, jq filters, file redirection. Assumes comfort with Unix pipelines.
🔴 Hard — multi-step pipelines, browser automation, human-in-the-loop. Treat these as templates to adapt.
Why these ten
Everything composes because of three rules: data → stdout, progress → stderr, JSON when piped. Your old friends jq, cron, and && do the rest.
For thirty years the deal with search engines was: you type, they list, you click, you copy. These recipes break that deal. A programmable search engine returns answers with citations, JSON shaped like your schema, or a finished browser session. The web stopped being a place you visit and became a standard library you call.