Make planner pipeline parallel-only

docs/cli/planner.md

@@ -16,11 +16,11 @@ lol plan --refine <issue-no> [refinement-instructions]
 
 1. **Understander** (sonnet) - Gathers codebase context with `Read,Grep,Glob` tools
 2. **Bold-proposer** (opus) - Researches SOTA solutions and proposes innovative approaches with `Read,Grep,Glob,WebSearch,WebFetch` tools and `--permission-mode plan`
-3. **Critique** (opus) - Validates assumptions and analyzes feasibility (runs in parallel with Reducer)
-4. **Reducer** (opus) - Simplifies proposal following "less is more" philosophy (runs in parallel with Critique)
+3. **Critique** (opus) - Validates assumptions and analyzes feasibility (always runs in parallel with Reducer)
+4. **Reducer** (opus) - Simplifies proposal following "less is more" philosophy (always runs in parallel with Critique)
 5. **Consensus** (opus) - Synthesizes final plan from the three reports using the external-consensus prompt
 
-Both critique and reducer append plan-guideline content and run in parallel via the Python executor.
+Critique and reducer append plan-guideline content and always run in parallel via the Python executor; there is no sequential mode.
 
 ### `--dry-run` (optional flag)
 

docs/feat/core/ultra-planner.md

@@ -65,7 +65,7 @@ graph TD
     U --> R{Lite conditions met?}
     R -->|yes: repo-only, <5 files, <150 LOC| L[Planner-lite: Single-agent plan]
     R -->|no: needs research or complex| C[Bold-proposer: Research SOTA]
-    C --> D[Critique + Reducer in parallel]
+    C --> D[Critique + Reducer (parallel-only)]
     D --> F[Combined 3-perspective report]
     L --> H[Update issue with plan]
     F --> G[External consensus: Synthesize plan]
@@ -258,7 +258,7 @@ After reviewing a plan issue:
 **Breakdown:**
 - Understander agent: 1-2 minutes (codebase exploration + complexity estimation)
 - Bold-proposer agent: 2-3 minutes (research + proposal, with context)
-- Critique + Reducer agents (parallel): 2-3 minutes
+- Critique + Reducer agents (parallel-only): 2-3 minutes
 - External consensus review: 1-2 minutes
 - Draft issue creation: <10 seconds
 
@@ -401,4 +401,3 @@ This does not change the `/ultra-planner` command interface documented above. Se
 | **Workflow** | Approval → Issue → Impl | Issue → Refine* → Impl |
 
 *Refinement is optional and can be done multiple times using `--refine`
-

python/agentize/workflow/README.md

@@ -37,7 +37,7 @@ and output suffix.
 ```
 understander → bold → critique → reducer → consensus (optional)
                       ↓           ↓
-                    (parallel when enabled)
+                     (parallel-only)
 ```
 
 1. **Understander**: Gathers codebase context and constraints
@@ -46,6 +46,8 @@ understander → bold → critique → reducer → consensus (optional)
 4. **Reducer**: Simplifies proposals following "less is more" philosophy
 5. **Consensus**: Synthesizes a unified implementation plan (optional for library use; CLI delegates to the external consensus script)
 
+Critique and reducer are always executed in parallel.
+
 ## Usage
 
 ```python
@@ -54,7 +56,6 @@ from agentize.workflow import run_planner_pipeline
 results = run_planner_pipeline(
     "Add user authentication with JWT tokens",
     output_dir=".tmp",
-    parallel=True,
     output_suffix="-output.md",
     skip_consensus=True,
 )

python/agentize/workflow/__init__.md

@@ -59,7 +59,6 @@ def run_planner_pipeline(
     *,
     output_dir: str | Path = ".tmp",
     backends: dict[str, tuple[str, str]] | None = None,
-    parallel: bool = True,
     runner: Callable[..., subprocess.CompletedProcess] = run_acw,
     prefix: str | None = None,
     output_suffix: str = "-output.md",
@@ -68,12 +67,12 @@ def run_planner_pipeline(
 ```
 
 Execute the 5-stage planner pipeline: understander → bold → critique → reducer → consensus.
+Critique and reducer are always executed in parallel.
 
 **Parameters:**
 - `feature_desc`: Feature request description to plan
 - `output_dir`: Directory for artifacts (default: `.tmp`)
 - `backends`: Provider/model mapping per stage (default: understander uses claude/sonnet, others claude/opus)
-- `parallel`: Run critique and reducer in parallel (default: True)
 - `runner`: Callable for stage execution (default: `run_acw`, injectable for testing)
 - `prefix`: Artifact filename prefix (default: timestamp-based)
 - `output_suffix`: Suffix appended to stage output filenames (default: `-output.md`)
@@ -162,7 +161,6 @@ from agentize.workflow import run_planner_pipeline
 results = run_planner_pipeline(
     "Implement dark mode toggle",
     backends={"consensus": ("claude", "opus")},
-    parallel=False,
 )
 
 for stage, result in results.items():

python/agentize/workflow/planner.md

@@ -12,7 +12,6 @@ def run_planner_pipeline(
     *,
     output_dir: str | Path = ".tmp",
     backends: dict[str, tuple[str, str]] | None = None,
-    parallel: bool = True,
     runner: Callable[..., subprocess.CompletedProcess] = run_acw,
     prefix: str | None = None,
     output_suffix: str = "-output.md",

python/agentize/workflow/planner/README.md

@@ -4,7 +4,7 @@ Runnable package for the multi-stage planner pipeline. Provides both library int
 
 ## Purpose
 
-This package contains the 5-stage planner pipeline (understander → bold → critique → reducer → consensus) that powers `lol plan`. It is structured as a runnable package to support `python -m agentize.workflow.planner` invocation, with the pipeline implemented as a Session DSL example.
+This package contains the 5-stage planner pipeline (understander → bold → critique → reducer → consensus) that powers `lol plan`. Critique and reducer are always executed in parallel. It is structured as a runnable package to support `python -m agentize.workflow.planner` invocation, with the pipeline implemented as a Session DSL example.
 
 ## Invocation
 
@@ -28,7 +28,6 @@ from agentize.workflow.planner import run_planner_pipeline, StageResult
 results = run_planner_pipeline(
     "Implement JWT authentication",
     output_dir=".tmp",
-    parallel=True,
 )
 
 for stage, result in results.items():

python/agentize/workflow/planner/__init__.md

@@ -12,7 +12,6 @@ def run_planner_pipeline(
     *,
     output_dir: str | Path = ".tmp",
     backends: dict[str, tuple[str, str]] | None = None,
-    parallel: bool = True,
     runner: Callable[..., subprocess.CompletedProcess] = run_acw,
     prefix: str | None = None,
     output_suffix: str = "-output.md",

python/agentize/workflow/planner/pipeline.md

@@ -12,7 +12,6 @@ def run_planner_pipeline(
     *,
     output_dir: str | Path = ".tmp",
     backends: dict[str, tuple[str, str]] | None = None,
-    parallel: bool = True,
     runner: Callable[..., subprocess.CompletedProcess] = run_acw,
     prefix: str | None = None,
     output_suffix: str = "-output.md",
@@ -64,7 +63,7 @@ artifacts (`*-consensus-input.md`, `*-consensus.md`).
 ## Design Rationale
 
 - **Session DSL as baseline**: The planner pipeline demonstrates the imperative Session API
-  with both sequential and parallel stages.
+  with a parallel-only critique/reducer stage.
 - **Explicit artifacts**: Stage-specific input/output files remain predictable and
   match CLI documentation.
 - **Reusable consensus stage**: Running consensus separately preserves the `.txt`

python/agentize/workflow/planner/pipeline.py

@@ -143,7 +143,6 @@ def run_planner_pipeline(
     *,
     output_dir: str | Path = ".tmp",
     backends: dict[str, tuple[str, str]] | None = None,
-    parallel: bool = True,
     runner: Callable[..., subprocess.CompletedProcess] = run_acw,
     prefix: str | None = None,
     output_suffix: str = "-output.md",
@@ -210,47 +209,30 @@ def _backend_label(stage: str) -> str:
         "reducer", feature_desc, agentize_home, bold_output
     )
 
-    mode_label = "parallel" if parallel else "sequentially"
     _log_stage(
-        f"Stage 3-4/5: Running critique and reducer {mode_label} "
+        "Stage 3-4/5: Running critique and reducer in parallel "
         f"({_backend_label('critique')}, {_backend_label('reducer')})"
     )
 
-    if parallel:
-        parallel_results = session.run_parallel(
-            [
-                session.stage(
-                    "critique",
-                    critique_prompt,
-                    stage_backends["critique"],
-                    tools=STAGE_TOOLS.get("critique"),
-                    permission_mode=STAGE_PERMISSION_MODE.get("critique"),
-                ),
-                session.stage(
-                    "reducer",
-                    reducer_prompt,
-                    stage_backends["reducer"],
-                    tools=STAGE_TOOLS.get("reducer"),
-                    permission_mode=STAGE_PERMISSION_MODE.get("reducer"),
-                ),
-            ]
-        )
-        results.update(parallel_results)
-    else:
-        results["critique"] = session.run_prompt(
-            "critique",
-            critique_prompt,
-            stage_backends["critique"],
-            tools=STAGE_TOOLS.get("critique"),
-            permission_mode=STAGE_PERMISSION_MODE.get("critique"),
-        )
-        results["reducer"] = session.run_prompt(
-            "reducer",
-            reducer_prompt,
-            stage_backends["reducer"],
-            tools=STAGE_TOOLS.get("reducer"),
-            permission_mode=STAGE_PERMISSION_MODE.get("reducer"),
-        )
+    parallel_results = session.run_parallel(
+        [
+            session.stage(
+                "critique",
+                critique_prompt,
+                stage_backends["critique"],
+                tools=STAGE_TOOLS.get("critique"),
+                permission_mode=STAGE_PERMISSION_MODE.get("critique"),
+            ),
+            session.stage(
+                "reducer",
+                reducer_prompt,
+                stage_backends["reducer"],
+                tools=STAGE_TOOLS.get("reducer"),
+                permission_mode=STAGE_PERMISSION_MODE.get("reducer"),
+            ),
+        ]
+    )
+    results.update(parallel_results)
 
     critique_output = results["critique"].text()
     reducer_output = results["reducer"].text()

python/tests/test_planner_workflow.py

@@ -168,37 +168,43 @@ class TestPlannerPipelineExecutionOrder:
     """Tests for correct stage execution order."""
 
     @pytest.mark.skipif(run_planner_pipeline is None, reason="Implementation not yet available")
-    def test_sequential_order_when_parallel_disabled(self, tmp_output_dir: Path, stub_runner: Callable):
-        """With parallel=False, stages run in deterministic order."""
+    def test_critique_reducer_run_parallel(self, tmp_output_dir: Path, stub_runner: Callable, monkeypatch):
+        """Critique and reducer are dispatched through the parallel runner."""
+        from agentize.workflow.planner import pipeline as planner_pipeline
+
+        recorded = {}
+
+        def _run_parallel(self, calls, *, max_workers: int = 2, retry: int = 0, retry_delay: float = 0):
+            call_list = list(calls)
+            recorded["stages"] = [call.stage for call in call_list]
+            results = {}
+            for call in call_list:
+                results[call.stage] = self.run_prompt(
+                    call.stage,
+                    call.prompt,
+                    call.backend,
+                    **call.options,
+                )
+            return results
+
+        monkeypatch.setattr(planner_pipeline.Session, "run_parallel", _run_parallel)
+
         run_planner_pipeline(
             "Add feature X",
             output_dir=tmp_output_dir,
             runner=stub_runner,
-            parallel=False,
             prefix="test",
         )
 
-        invocations = stub_runner.invocations
-        # Extract stage names from output file paths
-        stages = []
-        for inv in invocations:
-            output_path = inv["output_file"]
-            for stage in ["understander", "bold", "critique", "reducer", "consensus"]:
-                if stage in output_path:
-                    stages.append(stage)
-                    break
-
-        expected_order = ["understander", "bold", "critique", "reducer", "consensus"]
-        assert stages == expected_order
+        assert recorded.get("stages") == ["critique", "reducer"]
 
     @pytest.mark.skipif(run_planner_pipeline is None, reason="Implementation not yet available")
     def test_understander_runs_before_bold(self, tmp_output_dir: Path, stub_runner: Callable):
-        """Understander always runs before bold (even with parallel=True)."""
+        """Understander always runs before bold."""
         run_planner_pipeline(
             "Add feature Y",
             output_dir=tmp_output_dir,
             runner=stub_runner,
-            parallel=True,
             prefix="test",
         )
 

tests/CLAUDE.md

@@ -63,7 +63,7 @@ added to `test-all.sh` or executed directly. They provide shared functionality f
 
 ## Avoiding Embedded Python Blocks
 
-When writing shell tests, avoid embedding `python3 -c` blocks for testing Python logic. Instead:
+When writing shell tests, avoid embedding `python -c` blocks for testing Python logic. Instead:
 
 1. **Test Python logic in pytest**: Add test cases to `python/tests/test_*.py` files
 2. **Shell tests for CLI/integration**: Keep shell tests focused on CLI invocation, environment handling, and end-to-end workflows
@@ -72,7 +72,7 @@ When writing shell tests, avoid embedding `python3 -c` blocks for testing Python
 **Example - Avoid this in shell tests:**
 ```bash
 # BAD: Embedded Python block
-result=$(python3 -c "
+result=$(python -c "
 def some_function():
     return 'result'
 print(some_function())

tests/cli/test-cursor-hook-before-prompt-submit.sh

@@ -26,10 +26,10 @@ EOF
 )
 
     if [ -n "$agentize_home" ]; then
-        HANDSOFF_MODE="$handsoff_mode" AGENTIZE_HOME="$agentize_home" python3 "$HOOK_SCRIPT" <<< "$input"
+        HANDSOFF_MODE="$handsoff_mode" AGENTIZE_HOME="$agentize_home" python "$HOOK_SCRIPT" <<< "$input"
     else
         # Run without AGENTIZE_HOME (in local directory context)
-        (cd "$LOCAL_HOME" && unset AGENTIZE_HOME && HANDSOFF_MODE="$handsoff_mode" python3 "$HOOK_SCRIPT" <<< "$input")
+        (cd "$LOCAL_HOME" && unset AGENTIZE_HOME && HANDSOFF_MODE="$handsoff_mode" python "$HOOK_SCRIPT" <<< "$input")
     fi
 }
 

tests/cli/test-handsoff-session-path.sh

@@ -25,10 +25,10 @@ EOF
 )
 
     if [ -n "$agentize_home" ]; then
-        HANDSOFF_MODE=1 AGENTIZE_HOME="$agentize_home" python3 "$HOOK_SCRIPT" <<< "$input"
+        HANDSOFF_MODE=1 AGENTIZE_HOME="$agentize_home" python "$HOOK_SCRIPT" <<< "$input"
     else
         # Run without AGENTIZE_HOME (in local directory context)
-        (cd "$LOCAL_HOME" && unset AGENTIZE_HOME && HANDSOFF_MODE=1 python3 "$HOOK_SCRIPT" <<< "$input")
+        (cd "$LOCAL_HOME" && unset AGENTIZE_HOME && HANDSOFF_MODE=1 python "$HOOK_SCRIPT" <<< "$input")
     fi
 }