@@ -12,6 +12,11 @@
 # space left to move prose out.
 set -uo pipefail
 
+if ((BASH_VERSINFO[0] < 4)); then
+  echo "::error::check-workflow-size.sh requires Bash 4 or newer; found ${BASH_VERSION}. Run with a newer Bash (on macOS: brew install bash)." >&2
+  exit 1
+fi
+
 GITHUB_LIMIT_BYTES=512000
 GATE_BYTES="${WORKFLOW_SIZE_GATE_BYTES:-470000}"
 WARN_BYTES=$((GATE_BYTES - 25000))
 
@@ -70,7 +70,7 @@ if [ "$sandbox" = 'sandbox:docker' ]; then
     fi
     # Label- and age-filtered, so it cannot touch the image a
     # concurrent shard is testing.
-    docker image prune --all --force --filter 'label=org.qwen-code.ci.sandbox=true' --filter 'until=24h' || echo "::warning::old CI sandbox image cleanup failed on ${RUNNER_NAME:-this runner}"
+    timeout 20m docker image prune --all --force --filter 'label=org.qwen-code.ci.sandbox=true' --filter 'until=24h' || echo "::warning::old CI sandbox image cleanup failed on ${RUNNER_NAME:-this runner}"
     if ! build_image; then
       echo "::warning::sandbox image build failed; retrying once"
       build_image
 
@@ -0,0 +1,34 @@
+# E2E plan: H4e-b1 lead-side team runtime
+
+Scope: the Hosted lead's team tools (team_create, team_delete, task_create,
+task_update, task_list), the agent tool's `name`, the `<teammate>` label and
+the lead close cascade. The team domains ship disabled, so every run below
+needs a build with `team_state` and `team_task` added to
+`MANAGED_SESSION_ENABLED_DOMAINS` (the enablement step), on a real Hosted
+stack (managed-agent-server + qwen serve Harness + Runtime Broker).
+
+## Baseline (domains disabled, as shipped)
+
+1. Create a Hosted Session on `hosted-workspace-shell/1`; list the declared
+   tools of its first turn. Expect `agent` without `name`, and no team tool.
+2. Ask the model to launch an agent with `name`. Expect the tool error
+   "unsupported argument "name"" and no child_run record.
+
+## Physical acceptance (domains enabled)
+
+1. Lead creates a team, spawns two members by name, assigns board tasks,
+   receives both labeled reports (`<teammate>` in each notification), marks
+   the tasks done, and deletes the team. Expect: team_state chain opening →
+   2 joins → closing → deleted; task_list shows both members `completed`.
+2. Lead closes with a member still running. Expect: the member's child_run
+   ends `cancelled`/`stop_requested`, its Session closes, the team records
+   keep their last revision, and the lead is never revived by the member.
+3. Kill the Harness between a member's launch and its join (fault hook on the
+   `:join` commit) on a channel Turn, and restart. Expect exactly one
+   child_run, no roster entry, the child running on as an ordinary background
+   child, and the interrupted settlement answering the call as started. Repeat
+   with the hook after the join: expect the answer to say it joined.
+4. team_delete while a member runs. Expect the refusal naming it, and no
+   team_state revision.
+5. Approval in `default` mode: task_list runs without a card; team_create,
+   task_create and task_update cards show the input preview.
@@ -237,20 +237,25 @@ Read the diff and metadata, then write down — in the report — the PR's
 secondary claims. Budget by value:
 
 1. **A/B load-bearing proof of the central claim** (always, ~half the budget).
-   When the claim is about a turn's lifecycle (cancel, park, takeover,
-   recovery, approval, retry), name the scenario axes before choosing the
-   A/B instrument: take them from _Parameterise scenarios_ under wire-oracle
-   harnesses, and note the setting the author's tests use on each. The A/B
-   must also run at least one other setting, even when the instrument is the
-   PR's own test files: copy the author's new test and change only that
-   setting, such as parking Turn 2 after Turn 1 completed. The report lists
-   each axis with the settings that ran; the rest go under _Not covered_.
-   Measured example: a cancel fix settled a parked Turn only when it was the
-   Session's first. In a before/after re-run of this skill on that head,
-   neither arm named the axis. One ran the PR's own tests on base and head;
-   the other built cells that all parked the first Turn of a fresh Session.
-   The author's test, changed only to park Turn 2, failed there with
-   `expected 409 to be 200`.
+   When the diff changes how a Turn is cancelled, parked, taken over,
+   recovered, approved or retried, name the scenario axes before choosing
+   the A/B instrument, whether or not that code is the central claim; if it
+   is not, it gets its own A/B as a secondary claim. Take the axes from
+   _Parameterise scenarios_ under wire-oracle harnesses, and note the
+   setting the author's tests use on each. The A/B must also run at least
+   one other setting, even when the instrument is the PR's own test files:
+   copy the author's new test and change only that setting, such as parking
+   Turn 2 after Turn 1 completed. The report lists each axis with the
+   settings that ran; the rest go under _Not covered_. Measured example: a
+   cancel fix settled a parked Turn only when it was the Session's first.
+   In a before/after re-run of this skill on that head, neither arm named
+   the axis. One ran the PR's own tests on base and head; the other built
+   cells that all parked the first Turn of a fresh Session. The author's
+   test, changed only to park Turn 2, failed there with
+   `expected 409 to be 200`. A third arm had this rule keyed to the central
+   claim; it named Harness generation adoption as the claim, so the rule
+   never applied, although it mutated the cancellation-load branches. The
+   trigger is therefore the diff, not the claim.
 2. **One or two wire-oracle harnesses** on the changed surface.
 3. **Targeted gates**: tests/typecheck of the affected workspace(s) only.
 4. **Capture the A/B and the matrix as they print** — one command each,