auto

apsy:auto — smart router (natural language → /apsy:* command)

EXECUTION CONTRACT. Deterministic + transparent: a rule-based router (bin/apsy-route.py + config/routing.json) does the matching; the engine reports its reasons. Always show the user the chosen route and let them override before dispatch.

STEP 1 — Get the query

$ARGUMENTS is the user's intent. If empty, ask once via AskUserQuestion ("What do you want to do?") before continuing.

STEP 2 — Route

Run bin/apsy-route.py "<query>" — it auto-detects the nearest .apsy/state.json to apply a stage-aware boost (the current stage's next-action gets a bump). Parse the JSON: confidence, recommended_command, top (each with score + reasons), stage.

STEP 3 — Decide by confidence

• HIGH — tell the user "Routing to /apsy:<cmd> because ", then invoke the Skill apsy:<cmd> (or the matching one for the config commands like prolific/region/type).
• MEDIUM — present the top-2 candidates with AskUserQuestion (one option per candidate, plus "Other" for free text). Dispatch the chosen one.
• LOW — show the menu grouped by stage (FORMULATE / BUILD / PILOT / ANALYZE / PUBLISH / config-setters), ask AskUserQuestion which to run. Never auto-dispatch on LOW.

STEP 4 — Dispatch + record

Invoke the chosen apsy:<cmd> skill via the Skill tool. If we're inside an experiment directory, append a one-line routing record (query + chosen command + confidence) to .apsy/decisions.md for the audit trail.

PROHIBITED: silently routing on LOW confidence; routing to the G4-gated deploy without going through that skill's full G4 gate; inventing commands not in config/routing.json.

Validation gate: always print the chosen command (and the router's reasons) before invoking; never auto-dispatch on LOW.