whats-left

What’s left

One page that says where a project actually stands, and one questionnaire inside it that clears everything stopping it — answered in a sitting, exported as JSON, and read back so the answers turn into work.

Running as a Gemini model? Read gemini.md in this directory first, then follow this file with the overrides it names. It converts the Luke-voice pass into a phase that writes copy.json a later phase reads, puts a denominator on each survey scope, and reads back the four field bounds validate_model.py only warns about. Other models skip it.

The failure this exists to prevent

A status report and a list of open questions are usually two documents, and they drift the moment they are written. The report says a feature shipped; the question list asks whether to turn it on. Nobody can see that the second is why the first is a lie.

Two more failures follow from that split, and both are silent:

"Done" gets averaged. A feature that is written, merged and switched off in production is counted alongside one that clients used yesterday. DORA's own measures define deployed from deployment automation and never define done — so a page that reports one number for both is not simplifying, it is discarding the distinction that decides whether anyone should act.

A default gets read as an answer. A pre-selected recommendation is genuinely useful — it is how the reader gets through fourteen decisions in one sitting — but a default nobody looked at, exported as though it were chosen, is a decision attributed to a person who never made it. Every subsequent step inherits that as consent.

This skill produces one page in which neither can happen: the two halves share one link graph, stage is a word rather than a percentage, and the export distinguishes confirmed from the page proposed this and nobody looked.

When it runs

Produce mode — the default. "What's left before this is done", "what are you waiting on me for", "remaining work report", "/whats-left:whats-left".