v0.2 — Continuous Partner Capture

Phase 1 point release focused on how partner speech is captured.

Continuous partner capture

• The silence period (renamed Optional Responses silence period) is now a checkpoint, not a stop. When the partner pauses, the speech collected so far is sent for response options and a placeholder is spoken — but recording continues.
• If the partner keeps talking, the new speech is appended and the combined utterance is re-sent after the next pause, so each checkpoint yields a more complete set of options.
• Recording ends only when the user selects a response; the transcript is then cleaned once and the exchange stored.
• A latest-wins generation guard discards superseded in-flight requests so options never flicker backward.

New controls

• A persistent "Please repeat what you said." option, always visible, discards a garbled capture and keeps listening for the partner's restatement.
• A new auto-resume-listening setting (Settings → General; default off) turns partner recording back on automatically after a response is spoken.

Docs

• Architecture Overview §9 and the new Continuous Partner Capture memo document the revised flow.

Full changelog: 5c53d21...v0.2