spw.quest

spw.quest opens with a guide to the Spw language: one page to read, the twelve sigils, and the four brace pairs. This page is the guide beside it: constructs you can reuse, sensors you should run, and grammar you should not fork.

The install page puts the Spw Workbench into a repository. You choose who runs the setup, copy one block, run it from the repository root, and doctor tells you it worked. It stops before any commit.

Neighbor: texture.website keeps material cues and language texture next door. Quest is the path; texture is the grain.

Three blocks, one result. Each mounts the pinned workbench at .spw/_workbench, leaves your repository its own .spw/, and ends with doctor. Git and Node are the only things you bring.

For an agent reading this without a browser: spw.quest/init is the whole recipe as text, git.md is the five git commands in the order to run them, and quest.json is the same recipe as data. When it worked, spw:doctor finishes with no errors and nothing was committed; read the diff, then commit it yourself.

Paste this block into the agent you just gave the repository. Walk one path: consume the pinned workbench, or experiment through web semantics, not both in one patch. It is valid Spw, so you can check it before you trust it.

#>crawl_spwashi
?[what is this site doing that I could reuse]{
  open_first = #[`AGENTS.md`, `.spw/conventions/copy-accessor.spw`, `/about/domains/spw.quest/`, `/topics/software/spw/`]
  operation = `declare cache|audit|align|prime|contract|archive before editing`
  sense = `npm run sense -- copy|nouns|ink   ink needs a fixture id`
  person = `I'm Spwashi. I build software and make art.`
  dual_path = `consume pinned workbench XOR experiment through web semantics`
  write_deny = `Explore/plan do not write. Patch is one named slice.`
  braces = `[mode] keeps alternatives; (scene) stages midprocess — not decoration`
  wonder = `page types on body; harvest with npm run wonder — two inventories, not a backlog`
  report = `.spw/workbench-report.spw and /topics/software/spw/#workbench-improvements`
  walk = `Do not build a lens on an unpruned --from .spw walk`
  refuse = #[
    `fork Spw grammar`,
    `kitchen-sink implement from plans`,
    `fourth copy-accessor family`,
    `data-spw-wonder-type`,
    `git stash on a shared main`,
    `advertise physics the runtime does not spend`
  ]
  analog = `Spotify Portal: always-on markdown is I/O; a failing check is the block`
  synthesis = `If you can read X, cite a construct here. Mood is not a receipt.`
}

Source of record: .spw/caches/copy-agentic-direction-2026-09.spw#crawl_prompt. When the prompt changes there, it changes here.

This site is useful when a reading can be tried, checked, and refused. Opening { leans objective. Closing } leans subjective. The crawl is {mill}.{laminate}.{cure} — complete workshop practices. Tight {mill.laminate.cure} is one identifier the parser keeps whole, not a lifestyle list, and not cure-about-laminate-about-mill.

Check a reading with the literal parser, which runs the same pinned parser in the browser. Neighboring systems can keep .spw as sidecar curriculum, teaching beside their code without replacing the executable grammar. Since the 2026-09-03 pin, ; and || are real sequence marks; an or is still open.

Spw is useful to an agent when a claim can be tried, checked, and refused. The language is the map. The sensors are the block. A guide without a sensor is a suggestion.