feat: default compaction (256k) and max_depth=1

src/rlm/config.py

@@ -36,8 +36,13 @@ def _positive_int(value: str | int, name: str) -> int:
 
 
 def _summarize_at_tokens(value: str | int | None) -> int | None:
+    """Unset -> the 256k default; "" or "0" -> disabled; else a positive threshold."""
+    if value is None:
+        return 256_000
+    if value in ("", "0", 0):
+        return None
     parsed = _optional_positive_int(value, "summarize_at_tokens")
-    if value not in (None, "") and parsed is None:
+    if parsed is None:
         raise ValueError(f"summarize_at_tokens must be positive (got {value})")
     return parsed
 
@@ -124,10 +129,10 @@ def child(self) -> InvocationContext:
 class ExecutionPolicy(_ConfigModel):
     """Resource and context-management policy for one RLM engine."""
 
-    max_depth: int = Field(default=0, ge=0)
+    max_depth: int = Field(default=1, ge=0)
     exec_timeout: int = Field(default=300, gt=0)
     max_tokens: int | None = Field(default=None, gt=0)
-    summarize_at_tokens: int | None = Field(default=None, gt=0)
+    summarize_at_tokens: int | None = Field(default=256_000, gt=0)
     max_compactions: int | None = Field(default=None, gt=0)
     max_concurrent_subagents: int = Field(default=4, gt=0)
     max_subagent_calls: int = Field(default=64, gt=0)
@@ -161,7 +166,7 @@ def from_env(
     ) -> RuntimeConfig:
         env = os.environ if environ is None else environ
         raw_skills = env.get("RLM_SKILLS")
-        max_depth = int(env.get("RLM_MAX_DEPTH", "0"))
+        max_depth = int(env.get("RLM_MAX_DEPTH", "1"))
         default_concurrency = max(4, max_depth)
         max_concurrent_subagents = _positive_int(
             env.get("RLM_MAX_CONCURRENT_SUBAGENTS", str(default_concurrency)),

tests/test_acp.py

@@ -308,7 +308,7 @@ async def flaky_first_call(**kwargs):
         model="test-model",
         provider=ProviderConfig(base_url=None, api_key="test-key"),
         invocation=InvocationContext(),
-        policy=ExecutionPolicy(summarize_at_tokens=1),
+        policy=ExecutionPolicy(summarize_at_tokens=1, max_depth=0),
     )
     engine = RLMEngine(
         client=client,  # type: ignore[arg-type]

tests/test_config.py

@@ -64,3 +64,15 @@ def test_runtime_config_rejects_unsafe_recursive_and_environment_values():
     provider.headers["Idempotency-Key"] = "forged-after-construction"
     with pytest.raises(ValueError, match="reserved names"):
         make_client(provider)
+
+
+def test_default_policy_enables_compaction_and_recursion():
+    config = RuntimeConfig.from_env(environ={})
+    assert config.policy.summarize_at_tokens == 256_000
+    assert config.policy.max_depth == 1
+
+
+def test_summarize_at_tokens_disabled_by_zero_or_empty():
+    for raw in ("", "0"):
+        config = RuntimeConfig.from_env(environ={"RLM_SUMMARIZE_AT_TOKENS": raw})
+        assert config.policy.summarize_at_tokens is None