scene-capability-engine 3.6.48 → 3.6.50

package/template/.sce/README.md

@@ -20,7 +20,7 @@ This project uses **Spec-driven development** - a structured approach where:
 
 ---
 
-## 🚀 sce Capabilities (v1.45.x)
+## 🚀 sce Capabilities
 
 **IMPORTANT**: After installing or updating sce, read this section to understand all available capabilities. Using the right tool for the job ensures efficient, high-quality development.
 
@@ -63,7 +63,7 @@ This project uses **Spec-driven development** - a structured approach where:
 - Master Spec + Sub-Specs with dependency management
 - Interface contracts for cross-Spec compatibility
 
-### Multi-Agent Parallel Coordination (v1.43.0)
+### Multi-Agent Parallel Coordination
 When multiple AI agents work on the same project simultaneously:
 - **AgentRegistry** (`lib/collab`) — Agent lifecycle with heartbeat monitoring
 - **TaskLockManager** (`lib/lock`) — File-based task mutual exclusion
@@ -71,11 +71,11 @@ When multiple AI agents work on the same project simultaneously:
 - **SteeringFileLock** (`lib/lock`) — Steering file write serialization
 - **MergeCoordinator** (`lib/collab`) — Git branch isolation per agent
 - **Coordinator** (`lib/collab`) — Central task assignment (optional)
-- Config: `.sce/config/multi-agent.json` (`enabled: true` to activate)
-- All components are no-ops in single-agent mode (zero overhead)
+- Config: `.sce/config/multi-agent.json` (default `enabled: true`; set `enabled: false` to opt out)
+- If a project opts out with `enabled: false`, all components fall back to single-agent no-op behavior
 - See `docs/multi-agent-coordination-guide.md` for full API reference
 
-### Spec-Level Steering & Context Sync (v1.44.0)
+### Spec-Level Steering & Context Sync
 Fourth steering layer (L4) and Spec lifecycle coordination for multi-agent scenarios:
 - **SpecSteering** (`lib/steering`) — Per-Spec `steering.md` CRUD with template generation, Markdown ↔ structured object roundtrip
 - **SteeringLoader** (`lib/steering`) — Unified L1-L4 four-layer steering loader with merged output
@@ -83,7 +83,7 @@ Fourth steering layer (L4) and Spec lifecycle coordination for multi-agent scena
 - **SpecLifecycleManager** (`lib/collab`) — Spec state machine (planned → assigned → in-progress → completed → released) with auto-completion detection
 - **SyncBarrier** (`lib/collab`) — Agent Spec-switch synchronization barrier (uncommitted changes check, steering reload)
 - **Coordinator Integration** — `completeTask` auto-checks Spec completion; `assignTask` runs SyncBarrier
-- All components are no-ops in single-agent mode (zero overhead)
+- If needed, a project can still opt out by setting `enabled: false`, which restores single-agent no-op behavior
 - See `docs/multi-agent-coordination-guide.md` for full API reference
 
 ### Autonomous Control
@@ -92,7 +92,7 @@ Fourth steering layer (L4) and Spec lifecycle coordination for multi-agent scena
 - `sce auto status/resume/stop/config` — Manage autonomous execution
 - Intelligent error recovery, checkpoint system, learning from history
 
-### Agent Orchestrator — Multi-Agent Spec Execution (v1.45.0)
+### Agent Orchestrator — Multi-Agent Spec Execution
 Automate parallel Spec execution via Codex CLI sub-agents (replaces manual multi-terminal workflow):
 - `sce orchestrate run --specs "spec-a,spec-b,spec-c" --max-parallel 3` — Start multi-agent orchestration
 - `sce orchestrate status` — View orchestration progress (per-Spec status, overall state)
@@ -151,7 +151,7 @@ Look in .sce/specs/ directory
 
 ### When Working in Multi-Agent Mode
 
-If `.sce/config/multi-agent.json` exists with `enabled: true`:
+By default `.sce/config/multi-agent.json` is provisioned with `enabled: true`:
 1. Register with AgentRegistry before starting work
 2. Acquire task locks before modifying any task
 3. Use TaskStatusStore for concurrent-safe tasks.md updates
@@ -243,6 +243,6 @@ A Spec is a complete feature definition with three parts:
 ---
 
 **Project Type**: Spec-driven development  
-**sce Version**: 1.44.x  
-**Last Updated**: 2026-02-12  
+**sce Version**: 3.6.50  
+**Last Updated**: 2026-03-14  
 **Purpose**: Guide AI tools to work effectively with this project

package/template/.sce/config/multi-agent.json

@@ -0,0 +1,8 @@
+{
+  "enabled": true,
+  "heartbeatIntervalMs": 60000,
+  "heartbeatTimeoutMs": 180000,
+  "coordinatorEnabled": false,
+  "maxRetries": 5,
+  "retryBaseDelayMs": 100
+}

package/template/.sce/config/takeover-baseline.json

@@ -112,6 +112,17 @@
       "max_direct_fix_rounds_before_debug": 2,
       "forbid_bypass_workarounds": true
     },
+    "collaboration": {
+      "multi_user_mode": true,
+      "multi_agent": {
+        "enabled": true,
+        "heartbeatIntervalMs": 60000,
+        "heartbeatTimeoutMs": 180000,
+        "coordinatorEnabled": false,
+        "maxRetries": 5,
+        "retryBaseDelayMs": 100
+      }
+    },
     "migration_policy": {
       "legacy_kiro_supported": false,
       "require_manual_legacy_migration_confirmation": true

package/template/.sce/steering/CORE_PRINCIPLES.md

@@ -58,3 +58,11 @@
 - 前端调用后端 API 不匹配时，默认以后端现有接口契约为准。
 - 除非明确要求新建或修改后端接口，否则不要为了迁就前端错误调用去改后端。
 - 优先调整前端请求、映射、类型和兼容处理，使其与后端接口一致。
+
+## 11. 单文件规模过大时必须触发重构评估
+
+- SCE 应先评估当前项目的代码规模分布，再给出项目级的重构参考节点；不要把一个固定阈值生硬套到所有项目。
+- 如果项目还没有自己的阈值，默认参考源文件 `2000 / 4000 / 10000` 行三档触发：分别对应“必须评估”“必须发起重构收敛”“进入红线区”。
+- 达到项目级或默认阈值后，继续加功能前先评估是否应拆分职责；超过重构/红线阈值时，优先做拆分和降复杂度，而不是继续堆代码。
+- 项目早期应更早触发评估，后期也要按周或发布前复评，避免阈值长期失效。
+- 行数阈值只是强触发信号；如果文件虽然没到阈值，但职责混杂、测试困难、理解成本失控，也应提前重构。