Cover the silent-stall case (stacked on #12)

.claude/skills/shepherd/SKILL.md

@@ -477,6 +477,9 @@ from step 5 (Inner loop).
 - Verify runs on every run; the claim ledger is never empty.
 - Never report a dispatched stage as still running, and never end a turn waiting on one, without
   first checking its output file on disk — present and complete means done: read it and proceed.
+  An absent output file tells you only that the stage isn't done — not whether it's still
+  working or has died: report `status unknown; output not present` and offer to wait or
+  re-dispatch. Never infer "still running" or "stalled" from turn count or a human check-in.
 - Blindness per the role table's "do NOT read" column; judgment files are pasted, never granted;
   the repository itself is never blinded.
 - shepherd never commits `.shepherd/` paths. Run data stays ignored via the run's

tests/test_orchestrator_contract.py

@@ -201,10 +201,13 @@ def test_orchestrator_dispatches_reviewers_in_parallel():
 def test_dispatched_stage_completion_is_disk_based():
     # A dispatched stage's output file on disk is the completion signal, independent of
     # dispatch mode (background vs blocking) and dispatch site (single or parallel fan-out) —
-    # the orchestrator must never claim it's still waiting without checking disk first.
+    # the orchestrator must never claim it's still waiting without checking disk first. When the
+    # file is absent it reports the unknown honestly and never fabricates a status from turn count.
     assert "completion signal" in ORCH
     assert "output file on disk" in ORCH
     assert "Never report a dispatched stage as still running" in ORCH
+    assert "status unknown; output not present" in ORCH
+    assert 'Never infer "still running" or "stalled" from turn count or a human check-in' in ORCH
 
 
 def test_dispatched_stages_run_non_interactively():