docs: record OSA intake scheduler re-smoke

docs/internal/sessions/2026-05-27-osa-freebsd-intake-resmoke.md

@@ -0,0 +1,142 @@
+# OSA FreeBSD Intake Scheduler Re-smoke
+
+**Date:** 27.maj.2026  
+**Host:** osa.smilepowered.org  
+**OS:** FreeBSD 15.0-RELEASE-p9 amd64  
+**Repo:** `Clawdie/Colibri`  
+**Base pulled:** `af8c011` — daemon loop wiring landed  
+**Fix commit:** `d760536` — `fix: avoid scheduler store deadlock on intake drain`  
+**Status:** PASS after follow-up fix
+
+## What was checked
+
+Pulled the daemon-loop wiring (`9717ce7`, plus `af8c011`) and ran the full workspace gates:
+
+```sh
+cargo fmt --check
+cargo clippy --workspace --all-targets -- -D warnings
+cargo test --workspace
+cargo build --workspace --release
+```
+
+Initial gates were green, but a live `/tmp` FreeBSD re-smoke exposed one more runtime bug.
+
+## Finding during live re-smoke
+
+With `daemon::run_loop` now started, `intake-task` did reach the scheduler tick and inserted a SQLite task, but a concurrent `list-tasks` socket request timed out.
+
+Root cause: `Scheduler::tick` used expressions like:
+
+```rust
+match state.store.lock().unwrap().create_task(...) {
+    Ok(task) => {
+        state.store.lock().unwrap().list_agents()
+    }
+}
+```
+
+The `MutexGuard` temporary from the `match` scrutinee can live through the match arm. Relocking `state.store` inside the arm can deadlock the scheduler thread. On FreeBSD this manifested as:
+
+- SQLite row was created
+- socket request blocked on the store lock
+- daemon became unresponsive until killed by the smoke harness timeout
+
+## Fix
+
+`d760536` rewrites scheduler store access so every lock is scoped explicitly and dropped before the next lock:
+
+```rust
+let create_result = {
+    let store = state.store.lock().unwrap();
+    store.create_task(...)
+};
+```
+
+It also adds a regression test:
+
+```text
+scheduler::tests::test_scheduler_tick_drains_intake_without_deadlock
+```
+
+## Re-smoke result after fix
+
+Isolated `/tmp` environment:
+
+```text
+COLIBRI_DAEMON_DATA_DIR=/tmp/colibri-osa-resmoke-clawdie-1779908417/data
+COLIBRI_DAEMON_SOCKET=/tmp/colibri-osa-resmoke-clawdie-1779908417/colibri.sock
+COLIBRI_DB_PATH=/tmp/colibri-osa-resmoke-clawdie-1779908417/colibri.sqlite
+COLIBRI_HOST=osa-resmoke
+```
+
+`intake-task` response:
+
+```json
+{"ok":true,"data":{"status":"queued"}}
+```
+
+The scheduler drained intake on the 30s tick:
+
+```text
+FOUND_ON_POLL=30
+```
+
+`list-tasks` then returned the queued task:
+
+```json
+{
+  "ok": true,
+  "data": [
+    {
+      "agent_id": null,
+      "created_at": "2026-05-27T19:00:47.360062420+00:00",
+      "description": "prove scheduler loop drains intake",
+      "id": "c3dab9df-8a37-47b1-854b-d956fd796d41",
+      "status": "queued",
+      "title": "osa resmoke intake",
+      "updated_at": "2026-05-27T19:00:47.360062420+00:00"
+    }
+  ]
+}
+```
+
+SQLite verification:
+
+```text
+tasks 1
+journal_mode wal
+```
+
+Graceful shutdown:
+
+```text
+socket exists after stop? no
+process remains? no
+```
+
+Daemon log included both task loops exiting cleanly:
+
+```text
+daemon background loop started ... scheduler_secs=30
+received interrupt signal, initiating graceful shutdown
+socket server received shutdown signal
+daemon loop received shutdown signal
+daemon background loop exited
+Herdr socket API shut down
+colibri-daemon shut down cleanly
+```
+
+## Final verdict
+
+The daemon loop wiring is valid after `d760536`.
+
+Validated on FreeBSD:
+
+- daemon starts with `/tmp` data/socket/DB paths
+- background daemon loop starts beside the socket server
+- `intake-task` over Unix socket becomes a queued SQLite task on the scheduler tick
+- `list-tasks` remains responsive after the tick
+- SQLite WAL works
+- graceful shutdown removes socket and exits both socket + loop tasks
+
+The smoke directory was removed after recording this report.