fix codegen timeout

AGENTS.md

@@ -209,8 +209,11 @@ unit tests (24) + box integration tests (2).
   body…" badge) → sync codegen write → `materialize_skill` → hot-merge into the
   tree → the new leaf runs directly so the request is answered. `create_category`
   runs the same chain after authoring the category (`create_category` →
-  `create_skill` → run). Codegen failure leaves a navigable stub and returns a
-  graceful `create_skill` result.
+  `create_skill` → run). Codegen failure — including a request timeout — leaves
+  a navigable stub and returns a graceful `create_skill` result; a timeout is
+  raised as `CodegenError` by the client, never a raw `TimeoutError`. The
+  default codegen timeout is 1200s (`cli.build_scheduler`); raise
+  `codegen.timeout` in config for harder prompts.
 
 ### Code principles
 - **No mocking.** The decision engine is always real SemIf; the LLM is always a

semif_agent/cli.py

@@ -52,7 +52,7 @@ def build_scheduler(config: dict) -> tuple[Scheduler, dict]:
             "base_url", config.get("llm", {}).get("base_url", "http://localhost:11434/v1")
         ),
         model=codegen_cfg.get("model", "qwen38-iq3s"),
-        timeout=float(codegen_cfg.get("timeout", 600.0)),
+        timeout=float(codegen_cfg.get("timeout", 1200.0)),
     )
     log = DecisionLog(config.get("log", "data/decisions.jsonl"))
     trace = TraceLog(config.get("trace", "data/runs.jsonl"))

semif_agent/codegen.py

@@ -72,6 +72,10 @@ class CodegenClient:
             raise CodegenError(
                 f"codegen endpoint unreachable at {url}: {exc}. Is your local server running?"
             ) from exc
+        except TimeoutError as exc:
+            raise CodegenError(
+                f"codegen request timed out after {self.timeout}s at {url}"
+            ) from exc
         return payload["choices"][0]["message"]["content"]
 
 

tests/integration/test_pipeline.py

@@ -162,7 +162,7 @@ def test_generate_skill_body_codegen(tmp_path):
     client = CodegenClient(
         base_url=codegen_cfg.get("base_url", "http://localhost:11434/v1"),
         model=codegen_cfg.get("model", "qwen38-iq3s"),
-        timeout=float(codegen_cfg.get("timeout", 600.0)),
+        timeout=float(codegen_cfg.get("timeout", 1200.0)),
     )
     tree = build_tree(build_skills({"skills": {}}))
     draft = SkillDraft(
@@ -196,6 +196,7 @@ def test_create_skill_empty_category_does_not_wedge(tmp_path):
     config["log"] = str(tmp_path / "decisions.jsonl")
     config["trace"] = str(tmp_path / "runs.jsonl")
     scheduler, config = build_scheduler(config)
+    scheduler.codegen = None  # wedge regression only; skip the ~7min codegen body write
     scheduler.tree["travel_planning"] = []
 
     for _ in range(2):

tests/test_codegen.py

@@ -8,12 +8,14 @@ the CodegenClient itself is real, not mocked.
 
 import json
 import threading
+import time
 from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
 
 import pytest
 
 from semif_agent.codegen import (
     CodegenClient,
+    CodegenError,
     build_skill_body_prompt,
     generate_skill_body,
     parse_skill_body,
@@ -217,10 +219,39 @@ def test_codegen_client_unreachable_raises(tmp_path):
     client = CodegenClient(base_url="http://127.0.0.1:1/v1", model="test", timeout=2)
     tree = build_tree(build_skills({"skills": {}}))
     draft = SkillDraft(name="probe", description="Probe the service.")
-    with pytest.raises(Exception):
+    with pytest.raises(CodegenError):
         generate_skill_body(client, Request("is the service up?"), "tracking", draft, tree)
 
 
+class _SilentOpenAI(BaseHTTPRequestHandler):
+    """Accepts the request but never replies; the client must time out."""
+
+    def do_POST(self):
+        length = int(self.headers.get("Content-Length") or 0)
+        self.rfile.read(length)
+        time.sleep(5)
+
+    def log_message(self, format, *args):
+        pass
+
+
+def test_chat_timeout_raises_codegen_error():
+    handler = type("Handler", (_SilentOpenAI,), {})
+    httpd = ThreadingHTTPServer(("127.0.0.1", 0), handler)
+    threading.Thread(target=httpd.serve_forever, daemon=True).start()
+    try:
+        client = CodegenClient(
+            base_url=f"http://127.0.0.1:{httpd.server_address[1]}/v1",
+            model="test",
+            timeout=0.5,
+        )
+        with pytest.raises(CodegenError, match="timed out"):
+            client.chat([{"role": "user", "content": "hi"}])
+    finally:
+        httpd.shutdown()
+        httpd.server_close()
+
+
 def test_chat_omits_max_tokens_by_default():
     httpd, base = _fake_server(GOOD_BODY)
     try: