Measured examples
Nothing in the approved-workflow docs is documented until it has been run. These pages are the runs: real claude -p cells over the synthetic fixtures in test/workflows/fixtures/, each in its own clean box, recorded by plugins/arbeitsplan/scripts/subrun.py.
| case | cells | expected skill fired | caveat |
|---|---|---|---|
| Build a feature | 6 | 0/6 | see the page |
| Design a UI | 4 | 0/4 | see the page |
| Fix a bug | 6 | 5/6 | -- |
| Plan mode under an open run-scope lock | 4 | 4/4 | see the page |
| Review a plugin | 4 | 1/4 | -- |
| Which workflow fits? | 4 | 1/4 | -- |
| Understand an unfamiliar repo | 4 | 2/4 | -- |
Total recorded spend across these cells: $10.74. A cell that could not be measured fairly is scored UNMEASURED and excluded from every denominator; none of the cells here were.
Recorded 2026-09-14 from wf-2026-09-14-acceptedits, wf-2026-09-14-plan, wf-2026-09-14-plan-sonnet with 2.1.236 (Claude Code). Every number above comes from a cell record; regenerate with test/workflows/curate.py.