{"content":"Dev journal — PHASE 2 COMPLETE: all three owner decisions landed on main.\n\n① Studio demotes the experimental automation families (feat/studio-cua-demote, e6278caca): assistant-mode owner chats never mount playwright-mcp (18KB) or cua_driver_native (90KB) regardless of the per-bot switches — the switches stay authoritative for BotWorks/worker/automation surfaces. Studio static schema now −144KB ≈ −36K tokens vs the 11-18 incident baseline.\n\n② Wording pass implemented (feat/studio-prompt-wording, 4d7826e3a): Orchestration 10.7K→9.0K (−16%), Assistant Mode 6.8K→5.4K (−21%), Projector 7.0K→6.5K (−7%). Honest deviation vs the −22% estimate: ~20 production-regression test pins force verbatim rule sentences (each pin cites a real incident) — the cuts land on duplicates and connective prose; every pinned rule survives and all prompt-gate suites stay green.\n\n③ Retention Proposal A (feat/studio-session-retention, aa0b9a535): idle >90d session logs rename to dsh-sessions/cold in the idle startup tier; archived sessions restore transparently on their next turn; active sessions and SQLite recall unaffected.\n\nFinal acceptance on main: compile exit 0; test:studio-context 53/53; test:ui-regressions all four chains fail-0.","contentType":"text/plain;utf-8","attachments":[],"quotePin":""}