Review a workflow
Staves can criticise a board three ways, and count what it costs. They find different things, so the useful question is which one you want.
Findings — what is structurally wrong
Section titled “Findings — what is structurally wrong”what is wrong with this board?staves_issues and staves_review run structural and vocabulary checks. Typical findings:
- Jargon. A job named in system words a person in the workflow would never say.
- Coupled machinery. One job reached from two places, which usually means it is doing two jobs.
- No checks. A job that produces something and verifies nothing — nothing would tell a person it came out wrong.
- Nothing waiting. A job whose output nobody consumes.
These are cheap and worth running after every describe.
Reflection — what the shape implies
Section titled “Reflection — what the shape implies”reflect on this workflowstaves_reflect reads the board as a whole rather than job by job: unreachable jobs, dead ends,
nothing that returns to the outside party who asked, a must-not-happen rule that nobody guards,
intent that does not match where the work actually sits.
Hats — what each role would say
Section titled “Hats — what each role would say”review this as each role in turnstaves_hats takes every track in turn and comments on the jobs it touches, in that role’s voice
and interests. The supplier cares about being told why they are waiting; the budget owner cares
about being able to answer for a decision six months later. It surfaces the objections a single
reviewer does not think of, and it leaves the comments on the board rather than in chat.
Volume — what the shape costs
Section titled “Volume — what the shape costs”how often does each of these run?staves_volume records how many instances per week enter the board; per-job minutes come from
staves_describe. Together they turn “this step is slow” into “this step is slow four thousand
times a week”, which is usually what decides whether it is worth changing.
With a connected Langfuse project, staves_langfuse_runs and the local editor’s As run view show how the workflow actually ran in a window — runs, failures, exits, drift — measurement beside these reviews, not a substitute for them.
Questions beat opinions
Section titled “Questions beat opinions”Anything the review cannot determine should become a question on the board rather than a confident finding. You can answer them directly, and the answer stays attached to the job it concerns.