autoforge-ai 0.1.21 → 0.1.22

package/server/routers/projects.py

@@ -66,10 +66,12 @@ def _get_registry_functions():
         sys.path.insert(0, str(root))
 
     from registry import (
+        get_project_auto_improve,
         get_project_concurrency,
         get_project_path,
         list_registered_projects,
         register_project,
+        set_project_auto_improve,
         set_project_concurrency,
         unregister_project,
         validate_project_path,
@@ -82,6 +84,8 @@ def _get_registry_functions():
         validate_project_path,
         get_project_concurrency,
         set_project_concurrency,
+        get_project_auto_improve,
+        set_project_auto_improve,
     )
 
 
@@ -118,7 +122,7 @@ async def list_projects():
     _init_imports()
     assert _check_spec_exists is not None  # guaranteed by _init_imports()
     (_, _, _, list_registered_projects, validate_project_path,
-     get_project_concurrency, _) = _get_registry_functions()
+     get_project_concurrency, _, _, _) = _get_registry_functions()
 
     projects = list_registered_projects()
     result = []
@@ -140,6 +144,10 @@ async def list_projects():
             has_spec=has_spec,
             stats=stats,
             default_concurrency=info.get("default_concurrency", 3),
+            auto_improve_enabled=bool(info.get("auto_improve_enabled", False)),
+            auto_improve_interval_minutes=int(
+                info.get("auto_improve_interval_minutes", 10) or 10
+            ),
         ))
 
     return result
@@ -151,7 +159,7 @@ async def create_project(project: ProjectCreate):
     _init_imports()
     assert _scaffold_project_prompts is not None  # guaranteed by _init_imports()
     (register_project, _, get_project_path, list_registered_projects,
-     _, _, _) = _get_registry_functions()
+     _, _, _, _, _) = _get_registry_functions()
 
     name = validate_project_name(project.name)
     project_path = Path(project.path).resolve()
@@ -232,7 +240,8 @@ async def get_project(name: str):
     _init_imports()
     assert _check_spec_exists is not None  # guaranteed by _init_imports()
     assert _get_project_prompts_dir is not None  # guaranteed by _init_imports()
-    (_, _, get_project_path, _, _, get_project_concurrency, _) = _get_registry_functions()
+    (_, _, get_project_path, _, _, get_project_concurrency, _,
+     get_project_auto_improve, _) = _get_registry_functions()
 
     name = validate_project_name(name)
     project_dir = get_project_path(name)
@@ -246,6 +255,7 @@ async def get_project(name: str):
     has_spec = _check_spec_exists(project_dir)
     stats = get_project_stats(project_dir)
     prompts_dir = _get_project_prompts_dir(project_dir)
+    ai_enabled, ai_interval = get_project_auto_improve(name)
 
     return ProjectDetail(
         name=name,
@@ -254,6 +264,8 @@ async def get_project(name: str):
         stats=stats,
         prompts_dir=str(prompts_dir),
         default_concurrency=get_project_concurrency(name),
+        auto_improve_enabled=ai_enabled,
+        auto_improve_interval_minutes=ai_interval,
     )
 
 
@@ -267,7 +279,7 @@ async def delete_project(name: str, delete_files: bool = False):
         delete_files: If True, also delete the project directory and files
     """
     _init_imports()
-    (_, unregister_project, get_project_path, _, _, _, _) = _get_registry_functions()
+    (_, unregister_project, get_project_path, _, _, _, _, _, _) = _get_registry_functions()
 
     name = validate_project_name(name)
     project_dir = get_project_path(name)
@@ -304,7 +316,7 @@ async def get_project_prompts(name: str):
     """Get the content of project prompt files."""
     _init_imports()
     assert _get_project_prompts_dir is not None  # guaranteed by _init_imports()
-    (_, _, get_project_path, _, _, _, _) = _get_registry_functions()
+    (_, _, get_project_path, _, _, _, _, _, _) = _get_registry_functions()
 
     name = validate_project_name(name)
     project_dir = get_project_path(name)
@@ -338,7 +350,7 @@ async def update_project_prompts(name: str, prompts: ProjectPromptsUpdate):
     """Update project prompt files."""
     _init_imports()
     assert _get_project_prompts_dir is not None  # guaranteed by _init_imports()
-    (_, _, get_project_path, _, _, _, _) = _get_registry_functions()
+    (_, _, get_project_path, _, _, _, _, _, _) = _get_registry_functions()
 
     name = validate_project_name(name)
     project_dir = get_project_path(name)
@@ -368,7 +380,7 @@ async def update_project_prompts(name: str, prompts: ProjectPromptsUpdate):
 async def get_project_stats_endpoint(name: str):
     """Get current progress statistics for a project."""
     _init_imports()
-    (_, _, get_project_path, _, _, _, _) = _get_registry_functions()
+    (_, _, get_project_path, _, _, _, _, _, _) = _get_registry_functions()
 
     name = validate_project_name(name)
     project_dir = get_project_path(name)
@@ -395,7 +407,7 @@ async def reset_project(name: str, full_reset: bool = False):
         Dictionary with list of deleted files and reset type
     """
     _init_imports()
-    (_, _, get_project_path, _, _, _, _) = _get_registry_functions()
+    (_, _, get_project_path, _, _, _, _, _, _) = _get_registry_functions()
 
     name = validate_project_name(name)
     project_dir = get_project_path(name)
@@ -487,12 +499,13 @@ async def reset_project(name: str, full_reset: bool = False):
 
 @router.patch("/{name}/settings", response_model=ProjectDetail)
 async def update_project_settings(name: str, settings: ProjectSettingsUpdate):
-    """Update project-level settings (concurrency, etc.)."""
+    """Update project-level settings (concurrency, auto-improve, etc.)."""
     _init_imports()
     assert _check_spec_exists is not None  # guaranteed by _init_imports()
     assert _get_project_prompts_dir is not None  # guaranteed by _init_imports()
     (_, _, get_project_path, _, _, get_project_concurrency,
-     set_project_concurrency) = _get_registry_functions()
+     set_project_concurrency, get_project_auto_improve,
+     set_project_auto_improve) = _get_registry_functions()
 
     name = validate_project_name(name)
     project_dir = get_project_path(name)
@@ -509,10 +522,43 @@ async def update_project_settings(name: str, settings: ProjectSettingsUpdate):
         if not success:
             raise HTTPException(status_code=500, detail="Failed to update concurrency")
 
+    # Update auto-improve config if either field was provided
+    auto_improve_touched = (
+        settings.auto_improve_enabled is not None
+        or settings.auto_improve_interval_minutes is not None
+    )
+    if auto_improve_touched:
+        try:
+            success = set_project_auto_improve(
+                name,
+                enabled=settings.auto_improve_enabled,
+                interval_minutes=settings.auto_improve_interval_minutes,
+            )
+        except ValueError as e:
+            raise HTTPException(status_code=400, detail=str(e))
+        if not success:
+            raise HTTPException(status_code=500, detail="Failed to update auto-improve settings")
+
+        # Sync the scheduler with the new state.
+        from ..services.scheduler_service import get_scheduler
+        scheduler = get_scheduler()
+        ai_enabled, ai_interval = get_project_auto_improve(name)
+        if ai_enabled:
+            try:
+                await scheduler.register_auto_improve(name, project_dir, ai_interval)
+            except Exception as e:
+                raise HTTPException(
+                    status_code=500,
+                    detail=f"Failed to register auto-improve schedule: {e}",
+                )
+        else:
+            scheduler.remove_auto_improve(name)
+
     # Return updated project details
     has_spec = _check_spec_exists(project_dir)
     stats = get_project_stats(project_dir)
     prompts_dir = _get_project_prompts_dir(project_dir)
+    ai_enabled, ai_interval = get_project_auto_improve(name)
 
     return ProjectDetail(
         name=name,
@@ -521,4 +567,6 @@ async def update_project_settings(name: str, settings: ProjectSettingsUpdate):
         stats=stats,
         prompts_dir=str(prompts_dir),
         default_concurrency=get_project_concurrency(name),
+        auto_improve_enabled=ai_enabled,
+        auto_improve_interval_minutes=ai_interval,
     )

package/server/routers/settings.py

@@ -26,6 +26,7 @@ from registry import (
     AVAILABLE_MODELS,
     DEFAULT_MODEL,
     get_all_settings,
+    get_effort_setting,
     get_setting,
     set_setting,
 )
@@ -95,6 +96,8 @@ def _parse_bool(value: str | None, default: bool = False) -> bool:
     return value.lower() == "true"
 
 
+
+
 @router.get("", response_model=SettingsResponse)
 async def get_settings():
     """Get current global settings."""
@@ -114,6 +117,7 @@ async def get_settings():
         playwright_headless=True,  # Always headless - embedded browser view replaces desktop windows
         batch_size=_parse_int(all_settings.get("batch_size"), 3),
         testing_batch_size=_parse_int(all_settings.get("testing_batch_size"), 3),
+        effort=get_effort_setting(),
         api_provider=api_provider,
         api_base_url=all_settings.get("api_base_url"),
         api_has_auth_token=bool(all_settings.get("api_auth_token")),
@@ -142,6 +146,9 @@ async def update_settings(update: SettingsUpdate):
     if update.testing_batch_size is not None:
         set_setting("testing_batch_size", str(update.testing_batch_size))
 
+    if update.effort is not None:
+        set_setting("effort", update.effort)
+
     # API provider settings
     if update.api_provider is not None:
         old_provider = get_setting("api_provider", "claude")
@@ -182,6 +189,7 @@ async def update_settings(update: SettingsUpdate):
         playwright_headless=True,  # Always headless - embedded browser view replaces desktop windows
         batch_size=_parse_int(all_settings.get("batch_size"), 3),
         testing_batch_size=_parse_int(all_settings.get("testing_batch_size"), 3),
+        effort=get_effort_setting(),
         api_provider=api_provider,
         api_base_url=all_settings.get("api_base_url"),
         api_has_auth_token=bool(all_settings.get("api_auth_token")),

package/server/schemas.py

@@ -18,7 +18,7 @@ _root = Path(__file__).parent.parent
 if str(_root) not in sys.path:
     sys.path.insert(0, str(_root))
 
-from registry import DEFAULT_MODEL, VALID_MODELS
+from registry import DEFAULT_MODEL, LEGACY_MODEL_MAP, VALID_MODELS
 
 # ============================================================================
 # Project Schemas
@@ -46,6 +46,8 @@ class ProjectSummary(BaseModel):
     has_spec: bool
     stats: ProjectStats
     default_concurrency: int = 3
+    auto_improve_enabled: bool = False
+    auto_improve_interval_minutes: int = 10
 
 
 class ProjectDetail(BaseModel):
@@ -56,6 +58,8 @@ class ProjectDetail(BaseModel):
     stats: ProjectStats
     prompts_dir: str
     default_concurrency: int = 3
+    auto_improve_enabled: bool = False
+    auto_improve_interval_minutes: int = 10
 
 
 class ProjectPrompts(BaseModel):
@@ -75,6 +79,8 @@ class ProjectPromptsUpdate(BaseModel):
 class ProjectSettingsUpdate(BaseModel):
     """Request schema for updating project-level settings."""
     default_concurrency: int | None = None
+    auto_improve_enabled: bool | None = None
+    auto_improve_interval_minutes: int | None = None
 
     @field_validator('default_concurrency')
     @classmethod
@@ -83,6 +89,13 @@ class ProjectSettingsUpdate(BaseModel):
             raise ValueError("default_concurrency must be between 1 and 5")
         return v
 
+    @field_validator('auto_improve_interval_minutes')
+    @classmethod
+    def validate_auto_improve_interval(cls, v: int | None) -> int | None:
+        if v is not None and (v < 1 or v > 1440):
+            raise ValueError("auto_improve_interval_minutes must be between 1 and 1440")
+        return v
+
 
 # ============================================================================
 # Feature Schemas
@@ -471,6 +484,7 @@ class SettingsResponse(BaseModel):
     playwright_headless: bool = True
     batch_size: int = 3  # Features per coding agent batch (1-15)
     testing_batch_size: int = 3  # Features per testing agent batch (1-15)
+    effort: Literal["low", "medium", "high", "xhigh", "max"] = "xhigh"
     api_provider: str = "claude"
     api_base_url: str | None = None
     api_has_auth_token: bool = False  # Never expose actual token
@@ -491,6 +505,7 @@ class SettingsUpdate(BaseModel):
     playwright_headless: bool | None = None
     batch_size: int | None = None  # Features per agent batch (1-15)
     testing_batch_size: int | None = None  # Features per testing agent batch (1-15)
+    effort: Literal["low", "medium", "high", "xhigh", "max"] | None = None
     api_provider: str | None = None
     api_base_url: str | None = Field(None, max_length=500)
     api_auth_token: str | None = Field(None, max_length=500)  # Write-only, never returned
@@ -507,12 +522,16 @@ class SettingsUpdate(BaseModel):
 
     @field_validator('model')
     @classmethod
-    def validate_model(cls, v: str | None, info) -> str | None:  # type: ignore[override]
+    def validate_model(cls, v: str | None, info) -> str | None:
         if v is not None:
             # Skip VALID_MODELS check when using an alternative API provider
             api_provider = info.data.get("api_provider")
             if api_provider and api_provider != "claude":
                 return v
+            # Transparently accept legacy IDs so in-flight clients don't 422
+            # during an upgrade window; LEGACY_MODEL_MAP already covers migration.
+            if v in LEGACY_MODEL_MAP:
+                v = LEGACY_MODEL_MAP[v]
             if v not in VALID_MODELS:
                 raise ValueError(f"Invalid model. Must be one of: {VALID_MODELS}")
         return v

package/server/services/assistant_chat_session.py

@@ -270,8 +270,9 @@ class AssistantChatSession:
         system_cli = shutil.which("claude")
 
         # Build environment overrides for API configuration
-        from registry import DEFAULT_MODEL, get_effective_sdk_env
+        from registry import DEFAULT_MODEL, get_effective_sdk_env, get_effort_setting
         sdk_env = get_effective_sdk_env()
+        effort = get_effort_setting()
 
         # Determine model from SDK env (provider-aware) or fallback to env/default
         model = sdk_env.get("ANTHROPIC_DEFAULT_OPUS_MODEL") or os.getenv("ANTHROPIC_DEFAULT_OPUS_MODEL", DEFAULT_MODEL)
@@ -281,6 +282,7 @@ class AssistantChatSession:
             self.client = ClaudeSDKClient(
                 options=ClaudeAgentOptions(
                     model=model,
+                    effort=effort,  # type: ignore[arg-type]  # SDK 0.1.61 Literal omits "xhigh"
                     cli_path=system_cli,
                     # System prompt loaded from CLAUDE.md via setting_sources
                     # This avoids Windows command line length limit (~8191 chars)

package/server/services/expand_chat_session.py

@@ -161,8 +161,9 @@ class ExpandChatSession:
         system_prompt = skill_content.replace("$ARGUMENTS", project_path)
 
         # Build environment overrides for API configuration
-        from registry import DEFAULT_MODEL, get_effective_sdk_env
+        from registry import DEFAULT_MODEL, get_effective_sdk_env, get_effort_setting
         sdk_env = get_effective_sdk_env()
+        effort = get_effort_setting()
 
         # Determine model from SDK env (provider-aware) or fallback to env/default
         model = sdk_env.get("ANTHROPIC_DEFAULT_OPUS_MODEL") or os.getenv("ANTHROPIC_DEFAULT_OPUS_MODEL", DEFAULT_MODEL)
@@ -184,6 +185,7 @@ class ExpandChatSession:
             self.client = ClaudeSDKClient(
                 options=ClaudeAgentOptions(
                     model=model,
+                    effort=effort,  # type: ignore[arg-type]  # SDK 0.1.61 Literal omits "xhigh"
                     cli_path=system_cli,
                     system_prompt=system_prompt,
                     allowed_tools=[

package/server/services/process_manager.py

@@ -85,6 +85,7 @@ class AgentProcessManager:
         self.parallel_mode: bool = False  # Parallel execution mode
         self.max_concurrency: int | None = None  # Max concurrent agents
         self.testing_agent_ratio: int = 1  # Regression testing agents (0-3)
+        self.auto_improve: bool = False  # Auto-improve mode (single session)
 
         # Support multiple callbacks (for multiple WebSocket clients)
         self._output_callbacks: Set[Callable[[str], Awaitable[None]]] = set()
@@ -375,6 +376,7 @@ class AgentProcessManager:
         playwright_headless: bool = True,
         batch_size: int = 3,
         testing_batch_size: int = 3,
+        auto_improve: bool = False,
     ) -> tuple[bool, str]:
         """
         Start the agent as a subprocess.
@@ -386,6 +388,10 @@ class AgentProcessManager:
             max_concurrency: Max concurrent coding agents (1-5, default 1)
             testing_agent_ratio: Number of regression testing agents (0-3, default 1)
             playwright_headless: If True, run browser in headless mode
+            auto_improve: If True, run in auto-improve mode. Forces single-agent
+                execution (concurrency=1, testing_ratio=0) and passes
+                --auto-improve to the subprocess so it runs exactly one
+                improvement session and exits.
 
         Returns:
             Tuple of (success, message)
@@ -408,12 +414,19 @@ class AgentProcessManager:
         # Clean up features stuck from a previous crash/stop
         self._cleanup_stale_features()
 
+        # Auto-improve mode forces single-agent execution and skips testing
+        # agents — the subprocess bypasses the orchestrator entirely.
+        if auto_improve:
+            max_concurrency = 1
+            testing_agent_ratio = 0
+
         # Store for status queries
         self.yolo_mode = yolo_mode
         self.model = model
         self.parallel_mode = True  # Always True now (unified orchestrator)
         self.max_concurrency = max_concurrency or 1
         self.testing_agent_ratio = testing_agent_ratio
+        self.auto_improve = auto_improve
 
         # Build command - unified orchestrator with --concurrency
         cmd = [
@@ -432,6 +445,10 @@ class AgentProcessManager:
         if yolo_mode:
             cmd.append("--yolo")
 
+        # Add --auto-improve flag: bypasses the orchestrator for a one-shot run
+        if auto_improve:
+            cmd.append("--auto-improve")
+
         # Add --concurrency flag (unified orchestrator always uses this)
         cmd.extend(["--concurrency", str(max_concurrency or 1)])
 

package/server/services/scheduler_service.py

@@ -15,6 +15,7 @@ from typing import Optional
 
 from apscheduler.schedulers.asyncio import AsyncIOScheduler
 from apscheduler.triggers.cron import CronTrigger
+from apscheduler.triggers.interval import IntervalTrigger
 
 # Add parent directory for imports
 sys.path.insert(0, str(Path(__file__).parent.parent.parent))
@@ -73,19 +74,39 @@ class SchedulerService:
         logger.info("Scheduler service stopped")
 
     async def _load_all_schedules(self):
-        """Load schedules for all registered projects."""
+        """Load schedules and auto-improve jobs for all registered projects."""
         from registry import list_registered_projects
 
         try:
             projects = list_registered_projects()
             total_loaded = 0
+            total_auto_improve = 0
             for project_name, info in projects.items():
                 project_path = Path(info.get("path", ""))
-                if project_path.exists():
-                    count = await self._load_project_schedules(project_name, project_path)
-                    total_loaded += count
+                if not project_path.exists():
+                    continue
+
+                # Windowed schedules (cron-based)
+                count = await self._load_project_schedules(project_name, project_path)
+                total_loaded += count
+
+                # Auto-improve interval jobs (stored in the registry directly)
+                if info.get("auto_improve_enabled"):
+                    interval = int(info.get("auto_improve_interval_minutes", 10) or 10)
+                    try:
+                        await self.register_auto_improve(project_name, project_path, interval)
+                        total_auto_improve += 1
+                    except Exception as e:
+                        logger.error(
+                            f"Failed to register auto-improve for {project_name}: {e}"
+                        )
+
             if total_loaded > 0:
                 logger.info(f"Loaded {total_loaded} schedule(s) across all projects")
+            if total_auto_improve > 0:
+                logger.info(
+                    f"Registered {total_auto_improve} auto-improve job(s) across all projects"
+                )
         except Exception as e:
             logger.error(f"Error loading schedules: {e}")
 
@@ -205,6 +226,135 @@ class SchedulerService:
         else:
             logger.warning(f"No jobs found to remove for schedule {schedule_id}")
 
+    # ------------------------------------------------------------------
+    # Auto-improve interval jobs
+    # ------------------------------------------------------------------
+
+    @staticmethod
+    def _auto_improve_job_id(project_name: str) -> str:
+        return f"auto_improve_{project_name}"
+
+    async def register_auto_improve(
+        self,
+        project_name: str,
+        project_dir: Path,
+        interval_minutes: int,
+    ):
+        """Register or replace the auto-improve interval job for a project.
+
+        Uses APScheduler IntervalTrigger to fire every ``interval_minutes``.
+        On each tick, a single one-shot auto-improve agent session is started
+        (unless the project's agent is already running, in which case the tick
+        is silently skipped).
+        """
+        if interval_minutes < 1 or interval_minutes > 1440:
+            raise ValueError("interval_minutes must be between 1 and 1440")
+
+        job_id = self._auto_improve_job_id(project_name)
+        trigger = IntervalTrigger(minutes=interval_minutes)
+
+        self.scheduler.add_job(
+            self._handle_auto_improve_tick,
+            trigger,
+            id=job_id,
+            args=[project_name, str(project_dir)],
+            replace_existing=True,
+            misfire_grace_time=300,
+            max_instances=1,  # Never overlap ticks
+            coalesce=True,    # Collapse missed ticks during long runs
+        )
+
+        job = self.scheduler.get_job(job_id)
+        next_run = job.next_run_time if job else None
+        logger.info(
+            f"Registered auto-improve for {project_name}: "
+            f"every {interval_minutes} min (next: {next_run})"
+        )
+
+    def remove_auto_improve(self, project_name: str):
+        """Remove the auto-improve interval job for a project (no-op if missing)."""
+        job_id = self._auto_improve_job_id(project_name)
+        try:
+            self.scheduler.remove_job(job_id)
+            logger.info(f"Removed auto-improve job for {project_name}")
+        except Exception:
+            logger.debug(f"No auto-improve job to remove for {project_name}")
+
+    async def _handle_auto_improve_tick(
+        self,
+        project_name: str,
+        project_dir_str: str,
+    ):
+        """Fire one auto-improve agent session for a project.
+
+        Silently skips if the agent is already running (manual run, another
+        tick still executing, etc.). The scheduler's ``max_instances=1`` and
+        ``coalesce=True`` settings make sure ticks never stack up.
+        """
+        logger.info(f"Auto-improve tick for {project_name}")
+        project_dir = Path(project_dir_str)
+
+        if not project_dir.exists():
+            logger.warning(
+                f"Auto-improve tick: project dir missing for {project_name}, skipping"
+            )
+            return
+
+        try:
+            from .process_manager import get_manager
+
+            root_dir = Path(__file__).parent.parent.parent
+            manager = get_manager(project_name, project_dir, root_dir)
+
+            if manager.status in ("running", "paused", "pausing", "paused_graceful"):
+                logger.info(
+                    f"Auto-improve tick for {project_name}: agent already "
+                    f"{manager.status}, skipping this tick"
+                )
+                return
+
+            # Resolve effective yolo/model from global settings, mirroring the
+            # agent router's _get_settings_defaults() pattern so auto-improve
+            # respects whatever the user has configured globally.
+            yolo_mode, model = self._resolve_agent_defaults()
+
+            logger.info(
+                f"Starting auto-improve agent for {project_name} "
+                f"(yolo={yolo_mode}, model={model})"
+            )
+            success, msg = await manager.start(
+                yolo_mode=yolo_mode,
+                model=model,
+                max_concurrency=1,
+                testing_agent_ratio=0,
+                playwright_headless=True,
+                auto_improve=True,
+            )
+
+            if success:
+                logger.info(f"Auto-improve agent started for {project_name}")
+            else:
+                logger.warning(
+                    f"Auto-improve agent failed to start for {project_name}: {msg}"
+                )
+        except Exception as e:
+            logger.error(f"Error in auto-improve tick for {project_name}: {e}")
+
+    @staticmethod
+    def _resolve_agent_defaults() -> tuple[bool, str]:
+        """Resolve (yolo_mode, model) from global settings.
+
+        Kept separate from the agent router's helper so the scheduler never
+        has to import FastAPI routers. Mirrors the parsing behavior of
+        ``server/routers/agent.py::_get_settings_defaults``.
+        """
+        from registry import DEFAULT_MODEL, get_all_settings
+
+        settings = get_all_settings()
+        yolo_mode = (settings.get("yolo_mode") or "false").lower() == "true"
+        model = settings.get("api_model") or settings.get("model", DEFAULT_MODEL)
+        return yolo_mode, model
+
     async def _handle_scheduled_start(
         self, project_name: str, schedule_id: int, project_dir_str: str
     ):

package/server/services/spec_chat_session.py

@@ -147,8 +147,9 @@ class SpecChatSession:
         system_cli = shutil.which("claude")
 
         # Build environment overrides for API configuration
-        from registry import DEFAULT_MODEL, get_effective_sdk_env
+        from registry import DEFAULT_MODEL, get_effective_sdk_env, get_effort_setting
         sdk_env = get_effective_sdk_env()
+        effort = get_effort_setting()
 
         # Determine model from SDK env (provider-aware) or fallback to env/default
         model = sdk_env.get("ANTHROPIC_DEFAULT_OPUS_MODEL") or os.getenv("ANTHROPIC_DEFAULT_OPUS_MODEL", DEFAULT_MODEL)
@@ -157,6 +158,7 @@ class SpecChatSession:
             self.client = ClaudeSDKClient(
                 options=ClaudeAgentOptions(
                     model=model,
+                    effort=effort,  # type: ignore[arg-type]  # SDK 0.1.61 Literal omits "xhigh"
                     cli_path=system_cli,
                     # System prompt loaded from CLAUDE.md via setting_sources
                     # Include "user" for global skills and subagents from ~/.claude/