nextlisted

# /next — Run one iteration of the analysis loop Orchestrate one cycle of: **read state → accept-or-propose → write → run → propagate → close out**. Designed for projects past the initial data-pipeline stage where the workflow is "identify a question, write a focused script under `source/{table,figure}/`, inspect the build output, update the relevant docs." This is a *thin orchestrator*. Doc-writing steps invoke existing skills (`/findings`, `/hypothesis`, `/institutions`, `/check`, `/handoff`). The skill's job is the **sequence** and the **doc-propagation checklist**, not the writing of any one doc. ## Arguments - `/next <description>` — **specify mode** (primary). The researcher dictates what to do ("test whether H17 holds in the 2018+ subsample", "rebuild figure X with the post-fix sample"). Skip to step 2. - `/next` — **propose mode**. No description given. Scan project state and suggest 3–5 ranked candidates, then stop for greenlight. - `/next --project <slug>` — run against a specific project under `projects/`. Otherwise inferred from CWD. - `/next --propose-only` — stop after step 1; don't write or run anything. Useful for planning a session. - `/next --no-run` — write the script and stop. Useful when you want to inspect the script before it touches data. - `/next --auto` — **fully autonomous**. No stop-gates. If no description is given, propose internally, pick the top-ranked candidate, run it, propagate, close out. Use when you trust the proposal