fix: narrow broad exception handlers to prevent silent error swallowing

skillopt/engine/trainer.py

@@ -574,7 +574,7 @@ def _extract_failure_patterns(
                     ft = fs.get("failure_type", "")
                     sd = fs.get("description", "")
                     analyst_descs.append(f"{ft}: {sd}" if sd else ft)
-            except Exception:
+            except (json.JSONDecodeError, OSError, AttributeError, TypeError):
                 pass
 
     patterns = []

skillopt/envs/spreadsheetbench/adapter.py

@@ -113,7 +113,7 @@ def rollout(
                 for line in f:
                     try:
                         existing.append(json.loads(line))
-                    except Exception:
+                    except (json.JSONDecodeError, TypeError):
                         pass
             if existing:
                 return existing

skillopt/envs/spreadsheetbench/rollout.py

@@ -480,7 +480,7 @@ def run_spreadsheet_batch(
                     r = json.loads(line)
                     done_ids.add(str(r["id"]))
                     existing.append(r)
-                except Exception:
+                except (json.JSONDecodeError, KeyError, TypeError):
                     pass
 
     pending = [it for it in items if str(it["id"]) not in done_ids]
@@ -875,7 +875,7 @@ def run_spreadsheet_batch_codegen(
                     r = json.loads(line)
                     done_ids.add(str(r["id"]))
                     existing.append(r)
-                except Exception:
+                except (json.JSONDecodeError, KeyError, TypeError):
                     pass
 
     pending = [it for it in items if str(it["id"]) not in done_ids]

skillopt/gradient/aggregate.py

@@ -56,7 +56,7 @@ def _merge_batch(
             for e in merged.get(key, []):
                 e["merge_level"] = level
             return merged
-    except Exception:  # noqa: BLE001
+    except (ValueError, TypeError, KeyError, AttributeError, json.JSONDecodeError):
         pass
     # Fallback: concatenate all edits
     all_edits = []
@@ -248,7 +248,7 @@ def merge_patches(
                     f"{len(f_edits)}+{len(s_edits)} → {len(final[key])} {payload_label(update_mode)}"
                 )
             return final
-    except Exception:  # noqa: BLE001
+    except (ValueError, TypeError, KeyError, AttributeError, json.JSONDecodeError):
         pass
 
     return {

skillopt_sleep/backend.py

@@ -423,7 +423,7 @@ def judge(self, task: TaskRecord, response: str) -> Tuple[float, float, str]:
             try:
                 soft = float(obj.get("score", 0.0))
                 return (1.0 if soft >= 0.8 else 0.0), soft, str(obj.get("reason", ""))[:200]
-            except Exception:
+            except (ValueError, TypeError):
                 pass
         return 0.0, 0.0, "judge-parse-failed"
 
@@ -786,7 +786,7 @@ def _call(self, prompt: str, *, max_tokens: int = 1024) -> str:
             try:
                 import shutil
                 shutil.rmtree(clean_cwd, ignore_errors=True)
-            except Exception:
+            except OSError:
                 pass
         out = (proc.stdout or "").strip()
         self._detect_cli_error(out, proc.stderr or "")
@@ -861,7 +861,7 @@ def attempt_with_tools(self, task, skill, memory, tools):
         finally:
             try:
                 shutil.rmtree(work, ignore_errors=True)
-            except Exception:
+            except OSError:
                 pass
 
 def resolve_codex_path(explicit: str = "") -> str:
@@ -914,7 +914,7 @@ def resolve_codex_path(explicit: str = "") -> str:
             # skip the bash shim that execs hermes
             if head.startswith(b"#!") and b"bash" in head:
                 continue
-        except Exception:
+        except OSError:
             pass
         return c
     return "codex"
@@ -1001,7 +1001,7 @@ def _call_once(self, prompt: str, *, max_tokens: int = 1024) -> str:
         finally:
             try:
                 os.unlink(out_path)
-            except Exception:
+            except OSError:
                 pass
 
     # Fatal codex failures that will NOT recover on retry — fail fast + loud so a
@@ -1129,7 +1129,7 @@ def attempt_with_tools(self, task, skill, memory, tools):
             try:
                 with open(out_path, encoding="utf-8") as f:
                     resp = f.read().strip()
-            except Exception:
+            except OSError:
                 resp = ""
             # Surface a failed tool-rollout the SAME way _call does: an auth/model/version
             # failure on this path must show up in diagnostics (call_error), not vanish as a
@@ -1148,7 +1148,7 @@ def attempt_with_tools(self, task, skill, memory, tools):
         finally:
             try:
                 shutil.rmtree(work, ignore_errors=True)
-            except Exception:
+            except OSError:
                 pass
 
 def resolve_copilot_path(explicit: str = "") -> str:
@@ -1225,7 +1225,7 @@ def __init__(self, model: str = "", copilot_path: str = "", timeout: int = 240)
             )
             try:
                 os.makedirs(self.copilot_home, exist_ok=True)
-            except Exception:
+            except OSError:
                 self.copilot_home = ""
 
     def _call(self, prompt: str, *, max_tokens: int = 1024) -> str:
@@ -1260,7 +1260,7 @@ def _call(self, prompt: str, *, max_tokens: int = 1024) -> str:
             try:
                 import shutil
                 shutil.rmtree(clean_cwd, ignore_errors=True)
-            except Exception:
+            except OSError:
                 pass
         return self._parse_jsonl_response(proc.stdout or "")
 
@@ -1273,7 +1273,7 @@ def _parse_jsonl_response(raw: str) -> str:
                 continue
             try:
                 obj = json.loads(line)
-            except Exception:
+            except (json.JSONDecodeError, TypeError):
                 continue
             if obj.get("type") == "assistant.message":
                 content = (obj.get("data") or {}).get("content")
@@ -1380,7 +1380,7 @@ def attempt_with_tools(self, task, skill, memory, tools):
         finally:
             try:
                 shutil.rmtree(work, ignore_errors=True)
-            except Exception:
+            except OSError:
                 pass