diff --git a/.ai/HANDOFF.md b/.ai/HANDOFF.md
index 67716245..51872b94 100644
--- a/.ai/HANDOFF.md
+++ b/.ai/HANDOFF.md
@@ -8,11 +8,11 @@
 
 **Branch:** `fix/ci-workflow-config` → PR #150. PRs #151 and #152 were closed and consolidated into this branch — one PR is easier to land than three stacked ones, and the user got tired of waiting on serial CI runs of intermediate states.
 
-**Runner setup:** Three Gitea Actions agents are registered on the homelab box, so `backend` / `frontend` / `e2e` jobs run in parallel instead of serializing on a single agent. Combined with the xdist parallelization in this PR, the previous 1h 14m wall-clock should drop to ~9–12 min.
+**Runner setup:** Three Gitea Actions agents are registered on the homelab box, so `backend` / `frontend` / `e2e` jobs run truly in parallel. Combined with the xdist parallelization and the e2e decoupling in this PR, the previous 1h 14m wall-clock should drop to ~9 min.
 
 ## What's on PR #150's branch (consolidated)
 
-Eight commits, all CI-recovery work that landed together because they were inter-dependent:
+Seven CI-recovery commits that landed together because they were inter-dependent:
 
 1. **Codex's `49f8856 wip(handoff): restore backend suite to green`** — fixes the 54 real backend test failures left after #149.
 2. **Workflow correctness:** `DATABASE_TEST_URL` env, `actions/upload-artifact` v3 pin (Gitea Actions doesn't support v4+).
@@ -20,10 +20,21 @@ Eight commits, all CI-recovery work that landed together because they were inter
 4. **Postgres port-collision fix (`1bd43ab`):** dropped `ports: 5432:5432` host mapping. With three Gitea runner agents now active, two parallel jobs would race on `0.0.0.0:5432`. Tests connect via the `postgres` service-DNS hostname, not the host, so the mapping wasn't actually needed.
 5. **pytest-xdist with per-worker DBs (`7f71436`):** `pytest-xdist==3.6.1` added; `conftest.py` derives a per-worker DB URL from `PYTEST_XDIST_WORKER` and creates it synchronously on first import. Verified on PR #151's CI run before consolidation: 22m serial → 9m37s on the 4-core runner.
 6. **Five e2e selector updates (`69f2a37`):** drift from the FlowPilot/PSA migration. `Sessions` → `Session History`, `Account Settings` → `Account Management`, `/assistant` accepts `/pilot`, "Flow Sessions" tab clicks for ticket/client filtering and Resume on `/sessions`.
+7. **e2e decoupled from frontend (`261814a`):** dropped `needs: [frontend]` and the cross-job artifact handoff. e2e now builds its own frontend (npm ci + npm run build are already in the job). Adds ~1-2 min to the e2e job duration but removes the ~5 min of waiting for frontend to finish, and gets rid of the cross-job `actions/upload-artifact` mechanism entirely. e2e starts immediately on the third runner.
+
+## Expected wall-clock on the next CI run
+
+| Job | Duration | Starts at |
+|---|---|---|
+| backend | ~9 min (xdist on 4-core CI runner) | minute 0 |
+| frontend | ~6 min | minute 0 |
+| e2e | ~7 min (now self-builds) | minute 0 |
+
+Total wall-clock: ~9 min (whichever job runs longest), down from 1h 14m.
 
 ## Immediate next steps
 
-1. **Watch PR #150's CI** on its latest sha (`69f2a37`). Backend should run ~9 min via xdist; frontend ~6 min; e2e ~5 min with the selector fixes.
+1. **Watch PR #150's CI** on its latest sha (`261814a`). All three jobs should run concurrently and finish within ~9 min total.
 2. **Merge PR #150** when all three jobs are green.
 3. **Enable backend gate** on `main` branch protection — append `"CI / backend (pull_request)"` to `status_check_contexts`:
    ```bash