feat(rag): retry a failed LLM request before the first token lands — BOR_LLM_RETRIES/BOR_LLM_RETRY_DELAY with a live 'retrying' status

tests/unit/test_config.py

@@ -112,6 +112,41 @@ def test_max_output_tokens_env_override(monkeypatch) -> None:
     assert s.max_output_tokens == 1234
 
 
+def test_llm_retry_defaults(monkeypatch: pytest.MonkeyPatch) -> None:
+    """Phase 67: a failed LLM request is retried by default — 3 retries
+    with a flat 5 s delay (the TODO-locked values, no backoff)."""
+    monkeypatch.delenv("BOR_LLM_RETRIES", raising=False)
+    monkeypatch.delenv("BOR_LLM_RETRY_DELAY", raising=False)
+    s = _settings()
+    assert s.llm_retries == 3
+    assert s.llm_retry_delay == 5.0
+
+
+def test_llm_retry_env_overrides(monkeypatch: pytest.MonkeyPatch) -> None:
+    """``BOR_LLM_RETRIES`` / ``BOR_LLM_RETRY_DELAY`` override the defaults;
+    ``0`` retries is the no-retry kill switch (pre-phase-67 behavior)."""
+    monkeypatch.setenv("BOR_LLM_RETRIES", "0")
+    monkeypatch.setenv("BOR_LLM_RETRY_DELAY", "1.5")
+    s = _settings()
+    assert s.llm_retries == 0
+    assert s.llm_retry_delay == 1.5
+
+
+def test_llm_retries_rejects_negative(monkeypatch: pytest.MonkeyPatch) -> None:
+    """``0`` is the kill switch — a negative value is a typo, so the
+    validator fails loudly at startup (the ``agent_max_rounds`` pattern)."""
+    monkeypatch.setenv("BOR_LLM_RETRIES", "-1")
+    with pytest.raises(ValidationError, match="llm_retries"):
+        _settings()
+
+
+def test_llm_retry_delay_rejects_negative(monkeypatch: pytest.MonkeyPatch) -> None:
+    """A negative delay is a typo — fail loudly at startup."""
+    monkeypatch.setenv("BOR_LLM_RETRY_DELAY", "-0.5")
+    with pytest.raises(ValidationError, match="llm_retry_delay"):
+        _settings()
+
+
 def test_agent_max_rounds_default_and_env_override(monkeypatch: pytest.MonkeyPatch) -> None:
     """Phase 45: the per-tool budgets are gone — ``BOR_AGENT_MAX_ROUNDS``
     (default 10) is the single agent-loop knob; ``0`` is the no-tools