Fix checkpoint not updating on second resume

Root cause: checkpoint and tracer run directories used CWD-relative
paths (Path("strix_runs") and Path.cwd() / "strix_runs"). Launching
strix from different directories across sessions created separate
checkpoint files that never updated each other, so the third session
always resumed from the first session's iteration.

Fix: use Path.home() / "strix_runs" as the canonical absolute path in
both tracer.py and main.py so all sessions write to the same location
regardless of CWD.

Also includes earlier serialization robustness fixes (mode="json" +
_json_default fallback) and explicit checkpoint save in action_custom_quit.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

strix/interface/cli.py

@@ -314,7 +314,8 @@ async def run_cli(args: Any) -> None:  # noqa: PLR0915
     results_text = Text()
     results_text.append("Output", style="dim")
     results_text.append("  ")
-    results_text.append(f"strix_runs/{args.run_name}", style="#60a5fa")
+    from pathlib import Path
+    results_text.append(str(Path.home() / "strix_runs" / args.run_name), style="#60a5fa")
 
     note_text = Text()
     note_text.append("\n\n", style="dim")

strix/interface/main.py

@@ -606,20 +606,22 @@ def _setup_checkpoint_on_args(args: argparse.Namespace) -> None:
     args._checkpoint_data = None
     args.resume_from_checkpoint = False
 
+    strix_runs = Path.home() / "strix_runs"
+
     if args.force_new:
         # Delete any checkpoint for the current run name (if explicit) and start fresh
         if args.run_name:
-            run_dir = Path("strix_runs") / args.run_name
+            run_dir = strix_runs / args.run_name
             CheckpointManager(args.run_name, run_dir).delete()
         if not args.run_name:
             args.run_name = generate_run_name(args.targets_info)
-        run_dir = Path("strix_runs") / args.run_name
+        run_dir = strix_runs / args.run_name
         args._checkpoint_manager = CheckpointManager(args.run_name, run_dir)
         return
 
     # If no explicit run name, try auto-detect by target hash
     if not args.run_name:
-        found = _find_checkpoint_by_target_hash(Path("strix_runs"), target_hash)
+        found = _find_checkpoint_by_target_hash(strix_runs, target_hash)
         if found:
             run_name, checkpoint = found
             console = Console()
@@ -629,7 +631,7 @@ def _setup_checkpoint_on_args(args: argparse.Namespace) -> None:
                 f"[dim]Use --new to start fresh.[/]"
             )
             args.run_name = run_name
-            run_dir = Path("strix_runs") / run_name
+            run_dir = strix_runs / run_name
             args._checkpoint_manager = CheckpointManager(run_name, run_dir)
             args._checkpoint_data = checkpoint
             args.resume_from_checkpoint = True
@@ -638,7 +640,7 @@ def _setup_checkpoint_on_args(args: argparse.Namespace) -> None:
         args.run_name = generate_run_name(args.targets_info)
 
     # Explicit run name (or freshly generated) — look for its checkpoint
-    run_dir = Path("strix_runs") / args.run_name
+    run_dir = strix_runs / args.run_name
     mgr = CheckpointManager(args.run_name, run_dir)
     args._checkpoint_manager = mgr
 
@@ -727,7 +729,7 @@ def main() -> None:
         if tracer:
             posthog.end(tracer, exit_reason=exit_reason)
 
-    results_path = Path("strix_runs") / args.run_name
+    results_path = Path.home() / "strix_runs" / args.run_name
     display_completion_message(args, results_path)
 
     if args.non_interactive:

strix/interface/tui.py

@@ -2139,6 +2139,22 @@ class StrixTUIApp(App):  # type: ignore[misc]
 
             self._scan_thread.join(timeout=1.0)
 
+        # Explicitly save checkpoint here so it is always persisted before the
+        # TUI exits, regardless of whether atexit handlers run reliably after
+        # Textual's exit() call.
+        _mgr = self.agent_config.get("checkpoint_manager")
+        _agent = getattr(self, "_current_agent", None)
+        if _mgr and _agent:
+            import contextlib
+            with contextlib.suppress(Exception):
+                _mgr.save(
+                    _agent.state,
+                    self.tracer,
+                    self.scan_config,
+                    self.agent_config.get("target_hash", ""),
+                    _agent.max_iterations,
+                )
+
         self.tracer.cleanup()
 
         self.exit()

strix/telemetry/checkpoint.py

@@ -12,10 +12,21 @@ from datetime import UTC, datetime
 from pathlib import Path
 from typing import Any
 
+
 from pydantic import BaseModel, Field
 
 logger = logging.getLogger(__name__)
 
+
+def _json_default(obj: Any) -> Any:
+    """Fallback serialiser for objects json.dumps cannot handle natively."""
+    if hasattr(obj, "isoformat"):
+        return obj.isoformat()
+    if hasattr(obj, "model_dump"):
+        return obj.model_dump(mode="json")
+    return str(obj)
+
+
 CHECKPOINT_VERSION = "1.0"
 
 
@@ -106,7 +117,7 @@ class CheckpointManager:
             self.run_dir.mkdir(parents=True, exist_ok=True)
 
             state_dict: dict[str, Any] = (
-                agent_state.model_dump() if hasattr(agent_state, "model_dump") else {}
+                agent_state.model_dump(mode="json") if hasattr(agent_state, "model_dump") else {}
             )
 
             tracer_chat_messages: list[dict[str, Any]] = []
@@ -137,7 +148,7 @@ class CheckpointManager:
                     s = getattr(inst, "state", None)
                     if s is not None and s.parent_id is not None:
                         sub_agent_states[sid] = (
-                            s.model_dump() if hasattr(s, "model_dump") else {}
+                            s.model_dump(mode="json") if hasattr(s, "model_dump") else {}
                         )
             except Exception:  # noqa: BLE001
                 pass
@@ -157,8 +168,13 @@ class CheckpointManager:
                 sub_agent_states=sub_agent_states,
             )
 
-            # Atomic write: .tmp → rename
-            self._tmp_path.write_text(checkpoint.model_dump_json(indent=2), encoding="utf-8")
+            # Atomic write: .tmp → rename.  Use json.dumps with a fallback
+            # serialiser so non-standard objects (datetime, Pydantic models,
+            # etc.) never silently abort the save.
+            json_str = json.dumps(
+                checkpoint.model_dump(mode="json"), indent=2, default=_json_default
+            )
+            self._tmp_path.write_text(json_str, encoding="utf-8")
             os.rename(self._tmp_path, self.checkpoint_path)
 
         except Exception as e:  # noqa: BLE001

strix/telemetry/tracer.py

@@ -294,7 +294,7 @@ class Tracer:
 
     def get_run_dir(self) -> Path:
         if self._run_dir is None:
-            runs_dir = Path.cwd() / "strix_runs"
+            runs_dir = Path.home() / "strix_runs"
             runs_dir.mkdir(exist_ok=True)
 
             run_dir_name = self.run_name if self.run_name else self.run_id