fix: guard threadEmails.last against empty list

Add an isEmpty check before accessing threadEmails.last to prevent
StateError when emails are deleted between query and access.

Closes #341

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

.forgejo/workflows/monitor.yml

@@ -1,18 +0,0 @@
-name: Monitor Agent Loop
-
-on:
-  schedule:
-    - cron: '0 */2 * * *'  # every 2 hours
-  workflow_dispatch:
-
-jobs:
-  monitor:
-    name: Check Agent Loop Health
-    runs-on: ubuntu-latest
-    timeout-minutes: 5
-
-    steps:
-      - uses: actions/checkout@v4
-
-      - name: Check agent loop heartbeat
-        run: python3 scripts/agent_loop.py monitor

AGENTS.md

@@ -8,46 +8,41 @@ CLI tool `fgj` is available to query issues/PRs/actions.
 
 ## Issue Label Workflow
 
-We use issues, follow this label state machine:
+Automation is handled by [agentloop](https://github.com/guettli/agentloop) running every 5 minutes via cron. Add a label to trigger an agent:
 
-- **State/ToPlan** — Issue needs a plan written by an agent before implementation
-- **State/Planned** — Plan has been posted as a comment; awaiting human review
-- **State/Ready** — Issue is approved and ready for implementation
-- **State/InProgress** — Set while an agent (or human) is actively working
-- **State/Question** — Agent hit a blocker or needs clarification
+| Label | Trigger | Outcome |
+|---|---|---|
+| `loop/plan` | Planning agent reads the issue and writes an implementation plan as a comment | Issue moves to `loop/plan-done` |
+| `loop/code` | Coding agent implements the change, creates a branch + PR | Issue moves to `loop/code-done` |
 
-Full lifecycle:
+**State machine:**
 
 ```
-State/ToPlan → State/Planned   (automated: agent_loop.py runs a planning agent)
-State/Planned → State/Ready    (manual: human reviews the plan and approves)
-State/Ready → State/InProgress (automated: agent_loop.py before starting implementation)
-State/InProgress → closed      (automated: after PR is merged and CI passes)
-any state → State/Question     (automated or manual: when blocked)
+loop/plan  →  loop/plan-in-progress  →  loop/plan-done
+                                      ↘  NeedSupervisor  (on failure)
+
+loop/code  →  loop/code-in-progress  →  loop/code-done
+                                      ↘  NeedSupervisor  (on failure)
 ```
 
-List open issues ready to pick up:
+**Rules:**
+
+- Only issues authored by allowed users are picked up (guettli, guettlibot, guettlibot2, forgejo-actions).
+- An issue with `NeedSupervisor` needs human attention — investigate, fix, then re-label.
+- The coding agent opens a PR but does NOT close the issue. A human reviews the PR and closes the issue after merging.
+- Planning agents only post a comment — they do NOT write code or open PRs.
+- `loop/*` labels are managed by agentloop — do not set them manually while an agent is active.
+
+**Typical lifecycle for a new feature:**
 
-```bash
-fgj issue list --json --state open | jq '[.[] | select(.labels[].name == "State/Ready")] | .[] | {number, title, html_url}'
 ```
-
-Rules:
-
-- Never start implementation on an issue without `State/Ready`
-- Planning agents only post a plan comment — they do NOT write code or open PRs
-- After `State/Planned`, a human must review the plan and manually add `State/Ready`
-- When working via the agent loop: label transitions are set automatically
-  by `agent_loop.py` — do **not** set them yourself.
-- When working manually: switch to `State/InProgress` as your **first action**:
-  ```bash
-  fgj issue edit <NUMBER> --remove-label "State/Ready" --add-label "State/InProgress"
-  ```
-- If blocked, replace current state label with `State/Question` and leave a comment explaining the blocker
-- When done and CI is green, close the issue:
-  ```bash
-  fgj issue close <NUMBER>
-  ```
+1. Create issue
+2. Add label loop/plan   → agent writes plan as comment
+3. Review plan, request changes or approve
+4. Add label loop/code   → agent implements + opens PR
+5. Review PR, merge
+6. Close issue
+```
 
 ## Code conventions
 

DEVELOPMENT.md

@@ -188,5 +188,3 @@ Using SSH to `localhost` is preferred over complex X11/Wayland permission hacks.
 ## Daily Workflow
 
 Refer to the [README.md](./README.md#daily-workflow) for common development tasks and commands.
-
-<!-- agentloop code test passed -->

lib/data/repositories/email_repository_impl.dart

@@ -156,6 +156,7 @@ class EmailRepositoryImpl implements EmailRepository {
       return;
     }
 
+    if (threadEmails.isEmpty) return;
     final latest = threadEmails.last;
 
     // Collect unique participants across the whole thread.