latch-eval-tools 0.1.1__py3-none-any.whl → 0.1.4__py3-none-any.whl

latch_eval_tools/harness/claudecode.py

@@ -38,9 +38,9 @@ def run_claudecode_task(
         agent_log_file.unlink()
 
     enhanced_prompt = _enhance_prompt_with_local_files(task_prompt, work_dir)
-    enhanced_prompt += """
+    enhanced_prompt += f"""
 
-CRITICAL: You must write eval_answer.json BEFORE signaling completion.
+CRITICAL: You must write eval_answer.json BEFORE signaling completion in the {work_dir}.
 Correct order: 1) Perform analysis 2) Write eval_answer.json with your answer 3) Exit"""
 
     cmd = ["claude", "--print", "--dangerously-skip-permissions", "--verbose", "--output-format", "stream-json"]
@@ -59,6 +59,7 @@ Correct order: 1) Perform analysis 2) Write eval_answer.json with your answer 3)
             home_dir = Path.home()
             current_mode = home_dir.stat().st_mode
             home_dir.chmod(current_mode | stat.S_IXOTH)
+
             eval_cache_dir = home_dir / ".eval_cache"
             if eval_cache_dir.exists():
                 shutil.chown(eval_cache_dir, user="claude", group="claude")
@@ -67,6 +68,8 @@ Correct order: 1) Perform analysis 2) Write eval_answer.json with your answer 3)
                         shutil.chown(item, user="claude", group="claude")
                     except PermissionError:
                         pass
+
+            work_dir.chmod(0o777)
         except KeyError:
             run_as_claude_user = False
 

{latch_eval_tools-0.1.1.dist-info → latch_eval_tools-0.1.4.dist-info}/METADATA

@@ -1,11 +1,12 @@
 Metadata-Version: 2.4
 Name: latch-eval-tools
-Version: 0.1.1
+Version: 0.1.4
 Summary: Shared eval tools for single-cell bench, spatial bench, and future biology benchmarks.
 License-File: LICENSE
 Requires-Python: >=3.10
 Requires-Dist: aiohttp>=3.0.0
 Requires-Dist: anthropic>=0.72.0
+Requires-Dist: latch-config>=0.1.0
 Requires-Dist: latch>=2.0.0
 Requires-Dist: matplotlib>=3.0.0
 Requires-Dist: mini-swe-agent

{latch_eval_tools-0.1.1.dist-info → latch_eval_tools-0.1.4.dist-info}/RECORD

@@ -16,7 +16,7 @@ latch_eval_tools/graders/multiple_choice.py,sha256=HTyYWQGFv7ATmTNJMhUnisiuNb1F7
 latch_eval_tools/graders/numeric.py,sha256=Q19GtyYtatKRdPX9DlWJ8CKSQHUhJ6P6k7hnhZnKE04,6826
 latch_eval_tools/graders/spatial.py,sha256=8dNRsCVazCI98mHwhLHz0T-kyvLL4dTrsayu2DqjMdU,4258
 latch_eval_tools/harness/__init__.py,sha256=FyDDHRdEOUo845NIGqzF1CbOzaZfilsId8LxhtgEWHQ,764
-latch_eval_tools/harness/claudecode.py,sha256=gbbPtie4TALwkQ7zas5rPNtkL-SjA4yDVeDfoX1hWuY,7023
+latch_eval_tools/harness/claudecode.py,sha256=GtYHqF-To3zaHh2bpSJqWDmt_a7oikS0omf7DGtfdVs,7078
 latch_eval_tools/harness/minisweagent.py,sha256=SZwbnigvGOWU4tIgijJSyOr4urTe8n1Dgj6ViuL2uM4,9197
 latch_eval_tools/harness/plotsagent.py,sha256=Qw_q-6WzAsGo1RzKBzD0lzsXZeyGRGwfZ0AU5s_e00g,5017
 latch_eval_tools/harness/runner.py,sha256=f042LLhEAN4rvltPIhjSn588Z2AkP5tTk0pKDH69i2M,7264
@@ -26,8 +26,8 @@ latch_eval_tools/linter/explanations.py,sha256=Uzg-IH9YjZl__X7nF7lDc-4FSv619CE8L
 latch_eval_tools/linter/runner.py,sha256=zii4EbfsdWqdynzCxn1GLX0oC7rUgnrq0Vwsc-13YNk,4619
 latch_eval_tools/linter/schema.py,sha256=Acqs4VSwhOP4G85nUCghPquGMGQ413In7kAWaQ1BtNE,3688
 latch_eval_tools/linter/validators.py,sha256=iWcWBrcOmjv8abJEy0ch6Di2LbT5L8TVTe2cl88mQoU,21488
-latch_eval_tools-0.1.1.dist-info/METADATA,sha256=rY0HcZp3o4NILau2ZvgAYAOuBtbektHt37CgdD4OvDc,2852
-latch_eval_tools-0.1.1.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
-latch_eval_tools-0.1.1.dist-info/entry_points.txt,sha256=pxTKTp2rBe7xq_j1KSVZE6ULxDZv_BsbFjAGHEw7Tlo,66
-latch_eval_tools-0.1.1.dist-info/licenses/LICENSE,sha256=zepfPTPE3p7B8XbkQVZ-MeZwB9vmSoXQhMUvHNnlvQo,39
-latch_eval_tools-0.1.1.dist-info/RECORD,,
+latch_eval_tools-0.1.4.dist-info/METADATA,sha256=1RPdPFJksa7WLDEdxW97JGyqZ069mt524SdeB2GjaN4,2887
+latch_eval_tools-0.1.4.dist-info/WHEEL,sha256=WLgqFyCfm_KASv4WHyYy0P3pM_m7J5L9k2skdKLirC8,87
+latch_eval_tools-0.1.4.dist-info/entry_points.txt,sha256=pxTKTp2rBe7xq_j1KSVZE6ULxDZv_BsbFjAGHEw7Tlo,66
+latch_eval_tools-0.1.4.dist-info/licenses/LICENSE,sha256=zepfPTPE3p7B8XbkQVZ-MeZwB9vmSoXQhMUvHNnlvQo,39
+latch_eval_tools-0.1.4.dist-info/RECORD,,