Specs and the change workflow

A spec is implementation detail for an AI agent to build part of the system; WAD edits are decomposed into specs by an LLM.

A spec is not a diff — it is the worked-out detail an AI coding agent needs to implement *part of the system*: scope, instructions, acceptance criteria. The WAD stays the source of truth for what the system *is*; specs are the work packages that get it built or changed.

The change workflow: edit the WAD data to describe the target state and get it reviewed. Then an LLM (with a human shaping the cut — decomposition is hard to get right) reads the delta between the reviewed baseline and the new state — git diff, wcl diff <rev>:wad.wcl wad.wcl, or a wcl wad spec skeleton, which pins the baseline sha and the exact entity/field change list — and writes one or more specs sized for the implementing model.

How much decomposition depends on the implementer: a strong model can often implement directly from the WAD and the diff; smaller models need the change broken into narrow, explicit specs. Status tracks the work honestly: :planning → :in_progress → :complete (or :abandoned); the merged revision becomes the next baseline.