Your browser sucks, JavaScript disabled! Gunnar's console needs JavaScript and WebAssembly. Git over SSH on port 2222 does not.

nordisk/nornir

.nornir/autonom-for-idiots.md

autonom — the idiot guide

You want one rule: if the matrix is green, everything in the viz/CLI/MCP works. autonom makes that true by refusing to be green if any surface is untested. Here's how to live with it.

The one-liner

Green matrix ⟺ every button/command/RPC is wired and asserted. A new surface with no test fails the build — you can't forget one.

I added a new facett component / viz tab. What do I do?

Nothing special — just follow the contract and autonom finds it:

  1. impl facett_core::Facet for MyView (this is how autonom discovers it).
  2. Add caps(), state_json(), and register it in facett/tests/compliance.rs::registry().
  3. Give it local(...) and remote(...) constructors if it shows warehouse data (thin + fat).
  4. Write an inject-assert test (feed real data, assert state() — never just "didn't panic").

If you skip step 4, the gate goes red and tells you exactly which surface is bare. That's the point.

I started the viz and the Test tab is empty / says "not run yet"

That's correct now (AUT1): the matrix shows every expected repo × aspect with not-run-yet cells before you've run anything. To fill them:

nornir test all          # run the matrix across the workspace → results land in the warehouse

Reopen the viz: the not-run cells flip to green/red. (Thin mode reads them over Viz.TestResults.)

I edited code but didn't commit — is the matrix lying?

The release gate runs against your working tree, so it's honest. The thin-client viz shows a staleness badge ("indexed @ sha, N uncommitted changes — advisory") when your working tree differs from what the warehouse was indexed at. A commit SHA can't see uncommitted edits — autonom uses a working-tree content digest instead.

A test went red / a component "looks wrong". How do I fix it (no human)?

Follow the LAW: agent-ui-test-fixing.md — validate as DATA (read state_json + functional_status, never eyeball a screen), EMIT-DOCTRINE (turn "looks right" into numbers; render+Read a PNG only as a fallback), debug by binary search down the call chain, follow the data, and make the fix earn a fail-on-bug / pass-on-fix matrix test. The 🧪 Test pane links it ("How agents fix tests"); it ships in docs/book.pdf.

Where's the truth?