synthesiselisted

# /synthesise Read across a folder of markdown and write **one** file — `current-state.md` — that holds the reconciled state of what's known: corroborated findings with provenance, contradictions surfaced (not flattened), per-section staleness, and what changed since the last run. This is a **memory layer**, not a digest. The difference is the last point: it reconciles against its own previous output, so it compounds and self-corrects across runs. One-shot summarisers can't. **It is read-only to everything except its own output.** It never edits, moves, or restructures the source files. It imposes no taxonomy. It plugs into whatever the user already has. --- ## Invocation ``` /synthesise <folder> # synthesise the folder into one current-state.md /synthesise <folder> -- <hint> # hint steers focus (e.g. "this is one customer account") /synthesise <folder> --per-subfolder # one current-state.md per immediate subfolder (e.g. accounts/*) /synthesise <folder> --dry-run # print the synthesis to chat, do not write the file ``` - `<folder>` is the scope. `current-state.md` is written **into** that folder unless `--per-subfolder` is set. - Default scope is "this folder and everything under it, as one unit." - `--per-subfolder` treats each immediate child folder as its own unit (use for account/feature/person trees). --- ## The lens (optional, but the biggest quality lever) A synthesis is sharper when it knows *for whom and for what*. Th