fix: session resume, resource loading, and error path consistency

- don't advance resume cursor past failed must_complete tasks
- peek at last tool result before consuming (safe for retry)
- use Traversable.open() for zip/wheel compatibility
- return empty string (not None) for prompt on invalid globals

Author: anticomputer

src/seclab_taskflow_agent/available_tools.py

@@ -121,7 +121,7 @@ def _load(self, tooltype: AvailableToolType, toolname: str) -> DocumentModel:
                     f"Cannot load {toolname} because {pkg_dir} is not a valid directory."
                 )
             filepath = pkg_dir.joinpath(filename + ".yaml")
-            with open(filepath) as fh:
+            with filepath.open() as fh:
                 raw = yaml.safe_load(fh)
 
             # Validate header before full parse

src/seclab_taskflow_agent/prompt_parser.py

@@ -66,7 +66,7 @@ def parse_prompt_args(
         for g in args[0].globals:
             if "=" not in g:
                 logging.error(f"Invalid global variable format: {g}. Expected KEY=VALUE")
-                return None, None, None, None, None, help_msg
+                return None, None, None, None, "", help_msg
             key, value = g.split("=", 1)
             cli_globals[key.strip()] = value.strip()
 

src/seclab_taskflow_agent/runner.py

@@ -198,7 +198,7 @@ async def _build_prompts_to_run(
         if "result" not in task_prompt.lower():
             logging.warning("repeat_prompt enabled but no {{ result }} in prompt")
         try:
-            last_result = json.loads(last_mcp_tool_results.pop())
+            last_result = json.loads(last_mcp_tool_results[-1])
             text = last_result.get("text", "")
             try:
                 iterable_result = json.loads(text)
@@ -214,6 +214,9 @@ async def _build_prompts_to_run(
             logging.critical("No last MCP tool result available")
             raise
 
+        # Consume only after successful parse
+        last_mcp_tool_results.pop()
+
         if not iterable_result:
             await render_model_output("** 🤖❗MCP tool result iterable is empty!\n")
         else:
@@ -689,14 +692,6 @@ async def _deploy(ra: dict, pp: str) -> bool:
                     )
                     raise last_task_error
 
-                # Checkpoint after successful task
-                session.record_task(
-                    index=task_index,
-                    name=task_name,
-                    success=task_complete,
-                    tool_results=list(last_mcp_tool_results),
-                )
-
                 if must_complete and not task_complete:
                     logging.critical("Required task not completed ... aborting!")
                     await render_model_output("🤖💥 *Required task not completed ...\n")
@@ -707,6 +702,15 @@ async def _deploy(ra: dict, pp: str) -> bool:
                     )
                     break
 
+                # Checkpoint after task (must_complete failures break above
+                # without advancing the resume cursor)
+                session.record_task(
+                    index=task_index,
+                    name=task_name,
+                    success=task_complete,
+                    tool_results=list(last_mcp_tool_results),
+                )
+
         # All tasks completed successfully
         if session is not None and not session.error:
             session.mark_finished()