fix: don't truncate the raw response in StructuredOutputExtractionError

Real incident (2026-08-15, task b243dfe8): this exception's message used to truncate the failing response to [:2000] chars -- fine for the "process exited nonzero" error paths in this same file, where a short excerpt is enough, but actively counterproductive here: diagnosing a real production extraction failure with only the 2000-char preview, it was impossible to tell "the model's response was genuinely cut off mid-generation" from "this is just where the display limit happened to land" -- both look identical (text stopping mid-word) without the full text. No length cap now: this flows into task_execution._mark_blocked's str(exc), stored as a plain TEXT execution_event column with no practical size limit, so keeping the complete response is the honest, actually-diagnosable choice -- the same investigation that found this needed the full text and didn't have it. New test confirms a >2000-char failing response is preserved whole in the exception, not truncated. 20/20 in this file; every file today's fixes touch stays green (the only failures anywhere in the full suite are 15 pre-existing ones in a concurrent, unrelated task's own test file, confirmed identical before and after this change). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

fix: don't truncate the raw response in StructuredOutputExtractionError
Real incident (2026-08-15, task b243dfe8): this exception's message used to truncate the failing response to [:2000] chars -- fine for the "process exited nonzero" error paths in this same file, where a short excerpt is enough, but actively counterproductive here: diagnosing a real production extraction failure with only the 2000-char preview, it was impossible to tell "the model's response was genuinely cut off mid-generation" from "this is just where the display limit happened to land" -- both look identical (text stopping mid-word) without the full text. No length cap now: this flows into task_execution._mark_blocked's str(exc), stored as a plain TEXT execution_event column with no practical size limit, so keeping the complete response is the honest, actually-diagnosable choice -- the same investigation that found this needed the full text and didn't have it. New test confirms a >2000-char failing response is preserved whole in the exception, not truncated. 20/20 in this file; every file today's fixes touch stays green (the only failures anywhere in the full suite are 15 pre-existing ones in a concurrent, unrelated task's own test file, confirmed identical before and after this change). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
bboonnee
1 parent cac67c9a
Showing 2 changed files with 33 additions and 1 deletions
src/org_node_trigger.py
src/test_org_node_trigger.py
@@ -149,9 +149,22 @@ def trigger_org_node_structured(org_node, prompt, target_folder_path, output_sch
     required_keys = list(output_schema.get("required", []))
     extracted = extract_json_payload(payload.get("result", ""), required_keys=required_keys)
     if extracted is None:
+        # 2026-08-15 (real incident, task b243dfe8): this used to truncate
+        # to [:2000] chars -- fine for the "process exited nonzero" error
+        # paths above, where a short excerpt is enough context, but actively
+        # counterproductive here: the WHOLE POINT of this exception is
+        # diagnosing why extraction failed (genuine model-side truncation
+        # vs some other JSON-validity issue), and a truncated preview can
+        # LOOK like the response was cut off mid-generation when that's
+        # really just this display limit -- indistinguishable from the
+        # real thing without the full text, which is exactly what happened
+        # investigating this live. No length cap: this flows into
+        # task_execution._mark_blocked's str(exc), stored as a plain TEXT
+        # execution_event column with no practical size limit, so keeping
+        # the complete response is the honest, actually-diagnosable choice.
         raise StructuredOutputExtractionError(
             "could not extract a payload matching required keys {} from org_node={}'s response: {!r}".format(
-                required_keys, org_node, payload.get("result", "")[:2000]
+                required_keys, org_node, payload.get("result", "")
             )
         )
     return extracted
@@ -119,6 +119,25 @@ class TriggerOrgNodeStructuredTest(unittest.TestCase):
                     "validator", "check this", "/tmp/some-project", _SCHEMA
                 )
+    def test_extraction_error_preserves_the_full_response_not_truncated(self):
+        # Real incident (2026-08-15, task b243dfe8): this used to truncate
+        # the failing response to [:2000] chars in the exception message --
+        # diagnosing a real production failure with only that preview, it
+        # was impossible to tell "the model's response was genuinely cut
+        # off mid-generation" from "this is just where the 2000-char
+        # display limit landed" -- indistinguishable without the full text.
+        long_non_json_response = "not json at all, " * 200  # well over 2000 chars
+        self.assertGreater(len(long_non_json_response), 2000)
+        with patch("src.org_node_trigger.subprocess.run") as run:
+            run.return_value = _completed_process(
+                stdout_obj={"is_error": False, "result": long_non_json_response}
+            )
+            with self.assertRaises(org_node_trigger.StructuredOutputExtractionError) as ctx:
+                org_node_trigger.trigger_org_node_structured(
+                    "validator", "check this", "/tmp/some-project", _SCHEMA
+                )
+        self.assertIn(long_non_json_response, str(ctx.exception))
+
     def test_appends_json_fence_instruction_and_schema_to_prompt(self):
         with patch("src.org_node_trigger.subprocess.run") as run:
             run.return_value = _completed_process(