diff --git a/agents/runner.py b/agents/runner.py
index 0db3819..471b683 100644
--- a/agents/runner.py
+++ b/agents/runner.py
@@ -45,6 +45,12 @@ def _build_claude_env() -> dict:
     Also resolves ~/.nvm/versions/node/*/bin globs that launchctl may not expand.
     """
     env = os.environ.copy()
+
+    # Propagate canonical DB path to agent subprocesses so they don't fall back
+    # to module-relative path (which breaks in worktrees — KIN-129).
+    if "KIN_DB_PATH" not in env:
+        env["KIN_DB_PATH"] = str(Path.home() / ".kin" / "kin.db")
+
     existing = env.get("PATH", "").split(":")
 
     extra = list(_EXTRA_PATH_DIRS)
diff --git a/cli/main.py b/cli/main.py
index d00c9e7..6b29e65 100644
--- a/cli/main.py
+++ b/cli/main.py
@@ -616,83 +616,98 @@ def run_task(ctx, task_id, dry_run, allow_write):
     is_noninteractive = os.environ.get("KIN_NONINTERACTIVE") == "1"
     click.echo(f"Task: {task['id']} — {task['title']}")
 
-    # Step 1: PM decomposes
-    click.echo("Running PM to decompose task...")
-    pm_started_at = datetime.now(timezone.utc).isoformat()
-    pm_result = run_agent(
-        conn, "pm", task_id, project_id,
-        model="sonnet", dry_run=dry_run,
-        allow_write=allow_write, noninteractive=is_noninteractive,
-    )
-    pm_ended_at = datetime.now(timezone.utc).isoformat()
+    # Step 1a: Check for pre-built steps from revise endpoint (e.g. with analyst injection).
+    # Decision #866: steps built in web/api.py are saved as pending_steps and consumed here.
+    pending_steps = task.get("pending_steps")
+    if pending_steps:
+        pipeline_steps = pending_steps
+        models.update_task(conn, task_id, pending_steps=None)
+        click.echo(f"Using pre-built pipeline ({len(pipeline_steps)} steps, skipping PM)...")
+        pm_result = None
+        pm_started_at = pm_ended_at = None
+        if is_noninteractive:
+            click.echo("\n[non-interactive] Auto-executing pipeline...")
+        elif not click.confirm("\nExecute pipeline?"):
+            click.echo("Aborted.")
+            return
+    else:
+        # Step 1b: PM decomposes
+        click.echo("Running PM to decompose task...")
+        pm_started_at = datetime.now(timezone.utc).isoformat()
+        pm_result = run_agent(
+            conn, "pm", task_id, project_id,
+            model="sonnet", dry_run=dry_run,
+            allow_write=allow_write, noninteractive=is_noninteractive,
+        )
+        pm_ended_at = datetime.now(timezone.utc).isoformat()
 
-    if dry_run:
-        click.echo("\n--- PM Prompt (dry-run) ---")
-        click.echo(pm_result.get("prompt", "")[:2000])
-        click.echo("\n(Dry-run: PM would produce a pipeline JSON)")
-        return
+        if dry_run:
+            click.echo("\n--- PM Prompt (dry-run) ---")
+            click.echo(pm_result.get("prompt", "")[:2000])
+            click.echo("\n(Dry-run: PM would produce a pipeline JSON)")
+            return
 
-    if not pm_result["success"]:
-        click.echo(f"PM failed: {pm_result.get('output', 'unknown error')}", err=True)
-        raise SystemExit(1)
-
-    # Parse PM output for pipeline
-    output = pm_result.get("output")
-    if isinstance(output, str):
-        try:
-            output = json.loads(output)
-        except json.JSONDecodeError:
-            click.echo(f"PM returned non-JSON output:\n{output[:500]}", err=True)
+        if not pm_result["success"]:
+            click.echo(f"PM failed: {pm_result.get('output', 'unknown error')}", err=True)
             raise SystemExit(1)
 
-    if not isinstance(output, dict) or "pipeline" not in output:
-        click.echo(f"PM output missing 'pipeline' key:\n{json.dumps(output, indent=2)[:500]}", err=True)
-        raise SystemExit(1)
+        # Parse PM output for pipeline
+        output = pm_result.get("output")
+        if isinstance(output, str):
+            try:
+                output = json.loads(output)
+            except json.JSONDecodeError:
+                click.echo(f"PM returned non-JSON output:\n{output[:500]}", err=True)
+                raise SystemExit(1)
 
-    pipeline_steps = output["pipeline"]
-    if not isinstance(pipeline_steps, list) or not pipeline_steps:
-        click.echo(
-            f"PM returned empty or invalid pipeline: {pipeline_steps!r}", err=True
-        )
-        raise SystemExit(1)
-    analysis = output.get("analysis", "")
+        if not isinstance(output, dict) or "pipeline" not in output:
+            click.echo(f"PM output missing 'pipeline' key:\n{json.dumps(output, indent=2)[:500]}", err=True)
+            raise SystemExit(1)
 
-    # Save completion_mode from PM output to task (only if neither task nor project has explicit mode)
-    task_current = models.get_task(conn, task_id)
-    update_fields = {}
-    project = models.get_project(conn, project_id)
-    project_mode = project.get("execution_mode") if project else None
-    if not task_current.get("execution_mode") and not project_mode:
-        pm_completion_mode = models.validate_completion_mode(
-            output.get("completion_mode", "review")
-        )
-        update_fields["execution_mode"] = pm_completion_mode
-        import logging
-        logging.getLogger("kin").info(
-            "PM set completion_mode=%s for task %s", pm_completion_mode, task_id
-        )
+        pipeline_steps = output["pipeline"]
+        if not isinstance(pipeline_steps, list) or not pipeline_steps:
+            click.echo(
+                f"PM returned empty or invalid pipeline: {pipeline_steps!r}", err=True
+            )
+            raise SystemExit(1)
+        analysis = output.get("analysis", "")
 
-    # Save category from PM output (only if task has no category yet)
-    if not task_current.get("category"):
-        pm_category = output.get("category")
-        if pm_category and isinstance(pm_category, str):
-            pm_category = pm_category.upper()
-            if pm_category in models.TASK_CATEGORIES:
-                update_fields["category"] = pm_category
+        # Save completion_mode from PM output to task (only if neither task nor project has explicit mode)
+        task_current = models.get_task(conn, task_id)
+        update_fields = {}
+        project = models.get_project(conn, project_id)
+        project_mode = project.get("execution_mode") if project else None
+        if not task_current.get("execution_mode") and not project_mode:
+            pm_completion_mode = models.validate_completion_mode(
+                output.get("completion_mode", "review")
+            )
+            update_fields["execution_mode"] = pm_completion_mode
+            import logging
+            logging.getLogger("kin").info(
+                "PM set completion_mode=%s for task %s", pm_completion_mode, task_id
+            )
 
-    if update_fields:
-        models.update_task(conn, task_id, **update_fields)
+        # Save category from PM output (only if task has no category yet)
+        if not task_current.get("category"):
+            pm_category = output.get("category")
+            if pm_category and isinstance(pm_category, str):
+                pm_category = pm_category.upper()
+                if pm_category in models.TASK_CATEGORIES:
+                    update_fields["category"] = pm_category
 
-    click.echo(f"\nAnalysis: {analysis}")
-    click.echo(f"Pipeline ({len(pipeline_steps)} steps):")
-    for i, step in enumerate(pipeline_steps, 1):
-        click.echo(f"  {i}. {step['role']} ({step.get('model', 'sonnet')}): {step.get('brief', '')}")
+        if update_fields:
+            models.update_task(conn, task_id, **update_fields)
 
-    if is_noninteractive:
-        click.echo("\n[non-interactive] Auto-executing pipeline...")
-    elif not click.confirm("\nExecute pipeline?"):
-        click.echo("Aborted.")
-        return
+        click.echo(f"\nAnalysis: {analysis}")
+        click.echo(f"Pipeline ({len(pipeline_steps)} steps):")
+        for i, step in enumerate(pipeline_steps, 1):
+            click.echo(f"  {i}. {step['role']} ({step.get('model', 'sonnet')}): {step.get('brief', '')}")
+
+        if is_noninteractive:
+            click.echo("\n[non-interactive] Auto-executing pipeline...")
+        elif not click.confirm("\nExecute pipeline?"):
+            click.echo("Aborted.")
+            return
 
     # Step 2: Execute pipeline
     click.echo("\nExecuting pipeline...")
diff --git a/core/db.py b/core/db.py
index 51c41b6..5d8e47b 100644
--- a/core/db.py
+++ b/core/db.py
@@ -3,10 +3,11 @@ Kin — SQLite database schema and connection management.
 All tables from DESIGN.md section 3.5 State Management.
 """
 
+import os
 import sqlite3
 from pathlib import Path
 
-DB_PATH = Path(__file__).parent.parent / "kin.db"
+DB_PATH = Path(os.environ.get("KIN_DB_PATH") or (Path.home() / ".kin" / "kin.db"))
 
 SCHEMA = """
 -- Проекты (центральный реестр)
@@ -65,6 +66,7 @@ CREATE TABLE IF NOT EXISTS tasks (
     revise_comment TEXT,
     revise_count INTEGER DEFAULT 0,
     revise_target_role TEXT DEFAULT NULL,
+    pending_steps JSON DEFAULT NULL,
     labels JSON,
     category TEXT DEFAULT NULL,
     telegram_sent BOOLEAN DEFAULT 0,
@@ -400,6 +402,10 @@ def _migrate(conn: sqlite3.Connection):
         conn.execute("ALTER TABLE tasks ADD COLUMN revise_target_role TEXT DEFAULT NULL")
         conn.commit()
 
+    if "pending_steps" not in task_cols:
+        conn.execute("ALTER TABLE tasks ADD COLUMN pending_steps JSON DEFAULT NULL")
+        conn.commit()
+
     if "obsidian_vault_path" not in proj_cols:
         conn.execute("ALTER TABLE projects ADD COLUMN obsidian_vault_path TEXT")
         conn.commit()
diff --git a/core/models.py b/core/models.py
index 2553c47..d4694ce 100644
--- a/core/models.py
+++ b/core/models.py
@@ -37,6 +37,7 @@ _JSON_COLUMNS: frozenset[str] = frozenset({
     "tech_stack",
     "brief", "spec", "review", "test_result", "security_result", "labels",
     "smoke_test_result",
+    "pending_steps",
     "tags",
     "dependencies",
     "steps",
@@ -380,7 +381,7 @@ def update_task(conn: sqlite3.Connection, id: str, **fields) -> dict:
     """
     if not fields:
         return get_task(conn, id)
-    json_cols = ("brief", "spec", "review", "test_result", "security_result", "labels", "smoke_test_result")
+    json_cols = ("brief", "spec", "review", "test_result", "security_result", "labels", "smoke_test_result", "pending_steps")
     for key in json_cols:
         if key in fields:
             fields[key] = _json_encode(fields[key])
diff --git a/tests/test_kin_arch_023_regression.py b/tests/test_kin_arch_023_regression.py
new file mode 100644
index 0000000..6dae2df
--- /dev/null
+++ b/tests/test_kin_arch_023_regression.py
@@ -0,0 +1,228 @@
+"""Regression tests for KIN-ARCH-023 — Analyst injection into actual executed pipeline.
+
+Two concerns:
+1. At revise_count >= 2, analyst MUST be the first step in pending_steps saved to DB
+   (not just in the response body). The subprocess picks steps from pending_steps,
+   so a test that only checks the response body misses the real execution path.
+2. The condition simplification at api.py ~1056 is semantically correct:
+   `steps and (not steps or steps[0].get('role') != 'analyst')`
+   is equivalent to:
+   `steps and steps[0].get('role') != 'analyst'`
+   because when `steps` is truthy, `not steps` is always False.
+"""
+
+import pytest
+from unittest.mock import patch
+
+from core.db import init_db
+from core import models
+
+
+# ---------------------------------------------------------------------------
+# Fixtures
+# ---------------------------------------------------------------------------
+
+@pytest.fixture
+def api_client(tmp_path):
+    """FastAPI TestClient with isolated DB (same pattern as test_kin_128_regression.py)."""
+    import web.api as api_module
+    api_module.DB_PATH = tmp_path / "test.db"
+    from web.api import app
+    from fastapi.testclient import TestClient
+    client = TestClient(app)
+    client.post("/api/projects", json={"id": "p1", "name": "P1", "path": "/p1"})
+    client.post("/api/tasks", json={"project_id": "p1", "title": "Fix bug"})
+    return client
+
+
+@pytest.fixture
+def api_client_with_conn(tmp_path):
+    """Returns (TestClient, db_path) so tests can query DB directly."""
+    import web.api as api_module
+    db_path = tmp_path / "test.db"
+    api_module.DB_PATH = db_path
+    from web.api import app
+    from fastapi.testclient import TestClient
+    client = TestClient(app)
+    client.post("/api/projects", json={"id": "p1", "name": "P1", "path": "/p1"})
+    client.post("/api/tasks", json={"project_id": "p1", "title": "Fix bug"})
+    return client, db_path
+
+
+# ---------------------------------------------------------------------------
+# 1. Analyst actually persisted to pending_steps in DB (subprocess-visible path)
+# ---------------------------------------------------------------------------
+
+class TestAnalystInjectedIntoPendingSteps:
+    """The subprocess reads steps from pending_steps in DB, not from the HTTP response.
+    These tests verify the DB state, not just the response body.
+    """
+
+    def test_second_revise_saves_analyst_first_in_pending_steps(self, api_client_with_conn):
+        """revise_count=2: pending_steps[0].role == 'analyst' in DB after revise call."""
+        client, db_path = api_client_with_conn
+        steps = [{"role": "backend_dev", "model": "sonnet"}]
+
+        with patch("web.api._launch_pipeline_subprocess"):
+            client.post("/api/tasks/P1-001/revise", json={"comment": "r1", "steps": steps})
+            client.post("/api/tasks/P1-001/revise", json={"comment": "r2", "steps": steps})
+
+        conn = init_db(str(db_path))
+        task = models.get_task(conn, "P1-001")
+        conn.close()
+
+        pending = task.get("pending_steps") or []
+        assert pending, "pending_steps не должен быть пустым после 2-й ревизии"
+        assert pending[0].get("role") == "analyst", (
+            f"KIN-ARCH-023: analyst должен быть первым в pending_steps (subprocess читает именно их), "
+            f"получили: {pending[0].get('role')}"
+        )
+
+    def test_first_revise_does_not_add_analyst_to_pending_steps(self, api_client_with_conn):
+        """revise_count=1: analyst НЕ добавляется в pending_steps."""
+        client, db_path = api_client_with_conn
+        steps = [{"role": "backend_dev", "model": "sonnet"}]
+
+        with patch("web.api._launch_pipeline_subprocess"):
+            client.post("/api/tasks/P1-001/revise", json={"comment": "r1", "steps": steps})
+
+        conn = init_db(str(db_path))
+        task = models.get_task(conn, "P1-001")
+        conn.close()
+
+        pending = task.get("pending_steps") or []
+        assert not pending or pending[0].get("role") != "analyst", (
+            "analyst не должен быть первым шагом в pending_steps при revise_count=1"
+        )
+
+    def test_third_revise_saves_analyst_first_in_pending_steps(self, api_client_with_conn):
+        """revise_count=3: analyst также является первым в pending_steps."""
+        client, db_path = api_client_with_conn
+        steps = [{"role": "frontend_dev", "model": "sonnet"}]
+
+        with patch("web.api._launch_pipeline_subprocess"):
+            for comment in ("r1", "r2", "r3"):
+                client.post("/api/tasks/P1-001/revise", json={"comment": comment, "steps": steps})
+
+        conn = init_db(str(db_path))
+        task = models.get_task(conn, "P1-001")
+        conn.close()
+
+        pending = task.get("pending_steps") or []
+        assert pending, "pending_steps не должен быть пустым после 3-й ревизии"
+        assert pending[0].get("role") == "analyst", (
+            f"revise_count=3: первым в pending_steps должен быть analyst, получили: {pending[0].get('role')}"
+        )
+
+    def test_pending_steps_length_increases_by_one_after_analyst_injection(self, api_client_with_conn):
+        """После инжекции analyst длина pending_steps на 1 больше исходного списка шагов."""
+        client, db_path = api_client_with_conn
+        original_steps = [
+            {"role": "backend_dev", "model": "sonnet"},
+            {"role": "reviewer", "model": "sonnet"},
+        ]
+
+        with patch("web.api._launch_pipeline_subprocess"):
+            client.post("/api/tasks/P1-001/revise", json={"comment": "r1", "steps": original_steps})
+            client.post("/api/tasks/P1-001/revise", json={"comment": "r2", "steps": original_steps})
+
+        conn = init_db(str(db_path))
+        task = models.get_task(conn, "P1-001")
+        conn.close()
+
+        pending = task.get("pending_steps") or []
+        assert len(pending) == len(original_steps) + 1, (
+            f"Ожидали {len(original_steps) + 1} шагов (с analyst), получили {len(pending)}"
+        )
+
+    def test_analyst_not_duplicated_in_pending_steps_if_already_first(self, api_client_with_conn):
+        """Если analyst уже первый шаг — pending_steps не содержит дублей analyst."""
+        client, db_path = api_client_with_conn
+        steps = [{"role": "analyst", "model": "sonnet"}, {"role": "backend_dev", "model": "sonnet"}]
+
+        with patch("web.api._launch_pipeline_subprocess"):
+            client.post("/api/tasks/P1-001/revise", json={"comment": "r1", "steps": steps})
+            client.post("/api/tasks/P1-001/revise", json={"comment": "r2", "steps": steps})
+
+        conn = init_db(str(db_path))
+        task = models.get_task(conn, "P1-001")
+        conn.close()
+
+        pending = task.get("pending_steps") or []
+        analyst_count = sum(1 for s in pending if s.get("role") == "analyst")
+        assert analyst_count == 1, (
+            f"analyst не должен дублироваться в pending_steps, нашли {analyst_count} вхождений"
+        )
+
+
+# ---------------------------------------------------------------------------
+# 2. Condition simplification semantics (api.py ~1056)
+# ---------------------------------------------------------------------------
+
+def _original_condition(steps):
+    """Original (verbose) condition from before the simplification."""
+    return steps and (not steps or steps[0].get("role") != "analyst")
+
+
+def _simplified_condition(steps):
+    """Simplified condition after KIN-ARCH-023 fix."""
+    return steps and steps[0].get("role") != "analyst"
+
+
+class TestConditionSimplificationEquivalence:
+    """Verify that the simplified condition produces identical results to the original
+    across all meaningful input variants. This guards against future refactors
+    accidentally diverging the two forms.
+    """
+
+    def test_empty_list_both_return_falsy(self):
+        """steps=[] → оба варианта возвращают falsy (инжекция не происходит)."""
+        assert not _original_condition([])
+        assert not _simplified_condition([])
+
+    def test_none_both_return_falsy(self):
+        """steps=None → оба варианта возвращают falsy."""
+        assert not _original_condition(None)
+        assert not _simplified_condition(None)
+
+    def test_analyst_first_both_return_falsy(self):
+        """Если analyst уже первый — оба варианта возвращают falsy (инжекция пропускается)."""
+        steps = [{"role": "analyst"}, {"role": "backend_dev"}]
+        assert not _original_condition(steps)
+        assert not _simplified_condition(steps)
+
+    def test_non_analyst_first_both_return_truthy(self):
+        """Если первый шаг не analyst — оба варианта truthy (инжекция происходит)."""
+        steps = [{"role": "backend_dev"}, {"role": "reviewer"}]
+        assert _original_condition(steps)
+        assert _simplified_condition(steps)
+
+    def test_single_step_non_analyst_both_truthy(self):
+        """Одиночный шаг не-analyst → оба truthy."""
+        steps = [{"role": "debugger"}]
+        assert _original_condition(steps)
+        assert _simplified_condition(steps)
+
+    def test_step_without_role_key_both_truthy(self):
+        """Шаг без ключа 'role' → steps[0].get('role') возвращает None ≠ 'analyst' → truthy."""
+        steps = [{"model": "sonnet"}]
+        assert _original_condition(steps)
+        assert _simplified_condition(steps)
+
+    @pytest.mark.parametrize("steps", [
+        [],
+        None,
+        [{"role": "analyst"}],
+        [{"role": "backend_dev"}],
+        [{"role": "reviewer"}, {"role": "backend_dev"}],
+        [{"role": "analyst"}, {"role": "backend_dev"}],
+        [{"model": "sonnet"}],
+    ])
+    def test_parametrized_equivalence(self, steps):
+        """Для всех входных данных упрощённое условие идентично исходному."""
+        original = bool(_original_condition(steps))
+        simplified = bool(_simplified_condition(steps))
+        assert original == simplified, (
+            f"Условия расходятся для steps={steps}: "
+            f"original={original}, simplified={simplified}"
+        )
diff --git a/web/api.py b/web/api.py
index 9e2aa9b..53c481b 100644
--- a/web/api.py
+++ b/web/api.py
@@ -1053,7 +1053,7 @@ def revise_task(task_id: str, body: TaskRevise):
 
     # KIN-128: On 2nd+ revision, inject analyst as first step for fresh perspective.
     # Guard: skip if analyst is already the first step (idempotent), or if steps is None.
-    if revise_count >= 2 and steps and (not steps or steps[0].get("role") != "analyst"):
+    if revise_count >= 2 and steps and steps[0].get("role") != "analyst":
         analyst_step = {
             "role": "analyst",
             "model": "sonnet",
@@ -1065,6 +1065,11 @@ def revise_task(task_id: str, body: TaskRevise):
         }
         steps = [analyst_step] + list(steps)
 
+    # Persist computed steps so the subprocess can use them (avoids PM re-planning).
+    # Decision #866: modified pipeline steps must be saved to DB before subprocess launch.
+    if steps:
+        models.update_task(conn, task_id, pending_steps=steps)
+
     conn.close()
 
     # Launch pipeline in background subprocess