`8b1407b`

Clarify verification handoff

Authored by mfwolffe <wolffemf@dukes.jmu.edu> 1 week ago

SHA: 8b1407bbd52358cb2def7df0e12fc4de777de73b
Parents: d22c38c
Tree: 72f8922

4 changed files

Status	File	+	-
M	`src/loader/runtime/hooks.py`	2	2
M	`src/loader/runtime/parsing.py`	3	1
M	`src/loader/runtime/tool_batches.py`	29	20
M	`tests/test_tool_batches.py`	26	19

src/loader/runtime/hooks.pymodified

                          message=(
                              "[Blocked - post-build audit loop: all explicitly planned artifacts "
                              "already exist and the current output set has already been inspected "
 -                            "several times.] Suggestion: move to verification now or make one "
 -                            "concrete edit for a specific mismatch inside "
 +                            "several times.] Suggestion: finish with a final response so Loader "
 +                            "can verify automatically, or make one concrete edit for a specific mismatch inside "
                              f"{roots_preview} instead of more rereads."
                          ),
                          terminal_state="blocked",

src/loader/runtime/parsing.pymodified

                  if next_pending:
                      summary_parts.append(f"next pending: {next_pending}")
                  if payload.get("verification_nudge_needed") is True:
 -                    summary_parts.append("verification should be reviewed next")
 +                    summary_parts.append(
 +                        "final response should be provided next for Loader verification"
 +                    )
                  result = "; ".join(summary_parts)
      prefix = "Error" if is_error else "Result"

src/loader/runtime/tool_batches.pymodified

+ )
 +def _verification_handoff_instruction() -> str:
 +    return (
 +        "Finish with a final response now so Loader can run verification automatically. "
 +        "Do not run more ad hoc audit commands unless you know a specific mismatch to repair."
 +    )
++
++
  @dataclass
  class ToolBatchResult:
      """Outcome of running one assistant-proposed tool batch."""
          guidance = (
              "The remaining work is review/verification of the generated files. "
              "Do not ask the user for more clarification about the reference pattern now. "
 -            "Use the generated files as the source of truth and move to verification or "
 -            "repair from concrete failures in those files."
 +            "Use the generated files as the source of truth and repair any concrete failures "
 +            f"in those files. {_verification_handoff_instruction()}"
+         )
          if verification_commands:
              self.context.workflow_mode = "verify"
 -            guidance += " Verification should run next."
 +            guidance += " Do not run more ad hoc audit commands."
          self.context.queue_steering_message(guidance)
          message = Message.tool_result_message(
                  supplement_existing=True,
+             )
              verification_suffix = (
 -                "Move to verification or final confirmation using the files already on disk."
 +                _verification_handoff_instruction()
                  if verification_commands
                  else "Finish the current review using the files already on disk."
+             )
          if next_pending and _todo_is_consistency_review_step(next_pending):
              verification_suffix = (
 -                " If no specific mismatch remains, move to verification now."
 +                " If no specific mismatch remains, finish with a final response so Loader can verify."
                  if verification_commands
                  else " If no specific mismatch remains, finish the task now."
+             )
                  "All explicitly planned artifacts already exist. "
                  f"Use the generated files under {roots_preview} as the source of truth and stop broad rereads. "
                  "If you already know a concrete mismatch, fix it directly. "
 -                "Verification should run next. Do not reopen reference materials or keep auditing the same files."
 +                f"{_verification_handoff_instruction()} Do not reopen reference materials "
 +                "or keep auditing the same files."
+             )
              return
          verification_suffix = (
 -            "Move to verification or final confirmation using the files already on disk."
 +            _verification_handoff_instruction()
              if verification_commands
              else "Finish the task using the files already on disk."
+         )
                  f"with `{next_pending}` using the generated files as the source of truth. "
                  "Do not reopen earlier reference materials."
                  + (
 -                    " Verification should run next using those generated files."
 +                    " Finish with a final response so Loader can verify those generated files."
                      if verification_commands
                      else ""
+                 )
          self.context.queue_steering_message(
              "All explicitly planned artifacts already exist. "
              f"Stay within the current output roots under {roots_preview} "
 -            "and move to verification or final confirmation using the generated files. "
 +            "and finish with a final response so Loader can verify the generated files. "
              "Do not reopen earlier reference materials."
+         )
+         )
          if all_planned_artifacts_exist(dod, project_root=self.context.project_root):
              verification_suffix = (
 -                " Move to verification or final confirmation using the files already on disk."
 +                " " + _verification_handoff_instruction()
                  if verification_commands
                  else " If no concrete mismatch remains, stop editing and finish from the files already on disk."
+             )
                  supplement_existing=True,
+             )
              verification_suffix = (
 -                " Move to verification or final confirmation using the files already on disk."
 +                " " + _verification_handoff_instruction()
                  if verification_commands
                  else " Finish the task using the files already on disk."
+             )
              supplement_existing=True,
+         )
          verification_suffix = (
 -            " Move to verification or final confirmation using the files already on disk."
 +            " " + _verification_handoff_instruction()
              if verification_commands
              else " Finish the task using the files already on disk."
+         )
                  project_root=self.context.project_root,
+             )
          ):
 -            summary_parts.append("verification should be reviewed next")
 +            summary_parts.append("final response should be provided next for Loader verification")
          result = "; ".join(summary_parts)
          content = f"Observation [TodoWrite]: Result: {result}"
          if next_pending and _todo_is_consistency_review_step(next_pending):
              verification_suffix = (
 -                " Move to verification once no specific mismatch remains."
 +                " Finish with a final response once no specific mismatch remains so Loader can verify."
                  if verification_commands
                  else " Avoid another full reread unless one specific inconsistency is still unknown."
+             )
              self.context.queue_steering_message(
                  "All explicitly planned artifacts now exist on disk. "
                  "Do not expand the artifact set or restart discovery unless a specific gap is "
 -                "still known. Move to verification or final confirmation using the files that "
 -                "already exist."
 +                "still known. Finish with a final response now so Loader can verify the files "
 +                "that already exist."
+             )
      def _queue_missing_artifact_progress_nudge(
+             )
              if next_pending and _todo_is_consistency_review_step(next_pending):
                  verification_suffix = (
 -                    " Move to verification once no specific mismatch remains."
 +                    " Finish with a final response once no specific mismatch remains so Loader can verify."
                      if verification_commands
                      else " Finish the targeted consistency pass without reopening reference materials."
+                 )
                  self.context.set_workflow_mode("verify")
                  self.context.queue_steering_message(
                      "Todo tracking is updated. All explicitly planned artifacts now exist on disk. "
 -                    "Verification should run next. Use the current output files as the source of truth, "
 -                    "and do not restart discovery, reopen reference materials, or spend another turn "
 +                    "Finish with a final response now so Loader can run verification automatically. "
 +                    "Use the current output files as the source of truth, and do not restart discovery, "
 +                    "reopen reference materials, run more ad hoc audit commands, or spend another turn "
                      "on TodoWrite alone."
+                 )
                  return
              verification_suffix = (
 -                " Move to verification or final confirmation using the files already on disk."
 +                " " + _verification_handoff_instruction()
                  if verification_commands
                  else " Finish the task using the files already on disk."
+             )

tests/test_tool_batches.pymodified

      assert len(persistent_messages) == 1
      assert "All explicitly planned artifacts already exist on disk." in persistent_messages[0]
      assert (
 -        "Move to verification or final confirmation using the files already on disk."
 +        "Finish with a final response now so Loader can run verification automatically."
          in persistent_messages[0]
+     )
      assert "Create 07-performance-tuning.html" not in persistent_messages[0]
      assert len(persistent_messages) == 1
      assert "All explicitly planned artifacts already exist on disk." in persistent_messages[0]
      assert (
 -        "Move to verification or final confirmation using the files already on disk."
 +        "Finish with a final response now so Loader can run verification automatically."
          in persistent_messages[0]
+     )
      assert "Create 01-getting-started.html" not in persistent_messages[0]
      assert "Ensure all files are properly linked and formatted consistently" in message
      assert "Create 01-getting-started.html" not in message
      assert "do not keep broad-rereading the output set" in message
 -    assert "If no specific mismatch remains, move to verification now." in message
 +    assert "If no specific mismatch remains, finish with a final response so Loader can verify." in message
  @pytest.mark.asyncio
      assert len(persistent_messages) == 1
      assert "All explicitly planned artifacts already exist." in persistent_messages[0]
 -    assert "Verification should run next." in persistent_messages[0]
 +    assert "Finish with a final response now so Loader can run verification automatically." in persistent_messages[0]
      assert "stop broad rereads" in persistent_messages[0]
      assert ephemeral_messages == []
      assert context.workflow_mode == "verify"
          for message in persistent_messages
+     )
      assert any(
 -        "Move to verification once no specific mismatch remains." in message
 +        "Finish with a final response once no specific mismatch remains so Loader can verify."
 +        in message
          for message in persistent_messages
+     )
      message = queued_messages[-1]
      assert "Todo tracking is updated. All explicitly planned artifacts now exist on disk." in message
      assert "Verify all guide files are linked and complete" in message
 -    assert "Move to verification once no specific mismatch remains." in message
 +    assert (
 +        "Finish with a final response once no specific mismatch remains so Loader can verify."
 +        in message
 +    )
      assert "reopen reference materials" in message
      assert "Fortran guide structure" not in message
      assert context.workflow_mode == "execute"
      assert queued_messages
      message = queued_messages[-1]
      assert "Todo tracking is updated. All explicitly planned artifacts now exist on disk." in message
 -    assert "Verification should run next." in message
 +    assert "Finish with a final response now so Loader can run verification automatically." in message
      assert "Repair or verify the current files instead of expanding the artifact set." not in message
      assert context.workflow_mode == "verify"
      assert len(summary.tool_result_messages) == 1
      assert context.workflow_mode == "verify"
      assert queued_messages
 -    assert "Verification should run next." in queued_messages[-1]
 +    assert "Finish with a final response now so Loader can run verification automatically." in queued_messages[-1]
  @pytest.mark.asyncio
      assert result.continue_after_batch is True
      assert queued_messages
      message = queued_messages[-1]
 -    assert "Verification should run next." in message
 +    assert "Finish with a final response now so Loader can run verification automatically." in message
      assert "01-introduction.html" not in message
      assert "chapter files" not in message.lower()
      assert context.workflow_mode == "verify"
      assert summary.tool_result_messages
 -    assert "verification should be reviewed next" in summary.tool_result_messages[-1].content
 +    assert (
 +        "final response should be provided next for Loader verification"
 +        in summary.tool_result_messages[-1].content
 +    )
      assert "fortran guide structure" not in summary.tool_result_messages[-1].content.lower()
      assert [call.id for call in executor.calls] == ["bash-post-build-audit"]
      assert context.workflow_mode == "verify"
      assert queued_messages
 -    assert "Verification should run next." in queued_messages[-1]
 +    assert "Finish with a final response now so Loader can run verification automatically." in queued_messages[-1]
  @pytest.mark.asyncio
      assert queued_messages
      assert "The remaining work is review/verification of the generated files." in queued_messages[-1]
      assert "Do not ask the user for more clarification about the reference pattern now." in queued_messages[-1]
 -    assert "Verification should run next." in queued_messages[-1]
 +    assert "Finish with a final response now so Loader can run verification automatically." in queued_messages[-1]
      assert context.workflow_mode == "verify"
      assert summary.tool_result_messages
      assert "Skipped - stale post-build user question" in summary.tool_result_messages[-1].content
      assert summary.tool_result_messages
      message = summary.tool_result_messages[-1].content
      assert "updated todo list" in message
 -    assert "verification should be reviewed next" in message
 +    assert "final response should be provided next for Loader verification" in message
      assert "next pending:" not in message
      assert "fortran guide structure" not in message.lower()
      assert queued_messages
      message = queued_messages[-1]
      assert "Todo tracking is updated. All explicitly planned artifacts now exist on disk." in message
 -    assert "Verification should run next." in message
 +    assert "Finish with a final response now so Loader can run verification automatically." in message
      assert "Repair or verify the current files instead of expanding the artifact set." not in message
      assert "08-troubleshooting.html" not in message
      assert context.workflow_mode == "verify"
      assert queued
      assert "All explicitly planned artifacts already exist." in queued[0]
 -    assert "Move to verification or final confirmation using the files already on disk." in queued[0]
 +    assert "Finish with a final response now so Loader can run verification automatically." in queued[0]
      assert "replace the surrounding block" not in queued[0]
      assert "All explicitly planned artifacts already exist." in queued[0]
      assert "Verify all guide files are linked and complete" in queued[0]
      assert "Do not reopen earlier reference materials." in queued[0]
 -    assert "Verification should run next" in queued[0]
 +    assert "Finish with a final response so Loader can verify" in queued[0]
  def test_tool_batch_runner_blocked_post_build_audit_nudge_switches_to_verify(
      assert queued
      assert context.workflow_mode == "verify"
      assert "All explicitly planned artifacts already exist." in queued[0]
 -    assert "move to verification or final confirmation" in queued[0]
 +    assert "finish with a final response so Loader can verify" in queued[0]
  @pytest.mark.asyncio
      assert result.consecutive_errors == 0
      assert context.workflow_mode == "verify"
      assert queued
 -    assert any("move to verification or final confirmation" in message for message in queued)
 +    assert any("finish with a final response so Loader can verify" in message for message in queued)
  def test_tool_batch_runner_blocked_html_declared_target_nudge_uses_closest_declared_target(
      assert queued
      assert "All explicitly planned artifacts already exist on disk." in queued[0]
      assert "Do not expand the output set with `chapters/08-advanced-configuration.html`." in queued[0]
 -    assert "Move to verification or final confirmation using the files already on disk." in queued[0]
 +    assert "Finish with a final response now so Loader can run verification automatically." in queued[0]
      assert "update the guide root" not in queued[0]