@liangjie559567/ultrapower 5.6.6 → 5.6.8

package/commands/AGENTS.md

@@ -4,12 +4,13 @@
 # commands/
 
 ## Purpose
+
 包含 17 个斜杠命令定义（Axiom 工作流命令）。这些 Markdown 文件定义用户可通过 `/command-name` 调用的命令，与 `skills/` 目录中的 skill 实现保持镜像关系。
 
 ## Key Files
 
 | 文件 | 描述 |
-|------|------|
+| ------ | ------ |
 | `ax-draft.md` | `/ax-draft` — 需求起草工作流 |
 | `ax-review.md` | `/ax-review` — 5 专家并行评审 |
 | `ax-decompose.md` | `/ax-decompose` — 任务拆解为原子 DAG |
@@ -30,18 +31,25 @@
 ## For AI Agents
 
 ### Skills ↔ Commands 同步规则
-- `commands/` 中的每个命令文件对应 `skills/` 中的同名 skill 实现
-- 修改命令定义时，必须同步更新对应的 `skills/*/SKILL.md`
-- 两者功能描述必须保持一致
+
+* `commands/` 中的每个命令文件对应 `skills/` 中的同名 skill 实现
+
+* 修改命令定义时，必须同步更新对应的 `skills/*/SKILL.md`
+
+* 两者功能描述必须保持一致
 
 ### 修改此目录时
-- 新增命令需同时在 `skills/` 创建对应目录和 `SKILL.md`
-- 命令文件使用 Markdown 格式，包含触发词、描述和使用示例
+
+* 新增命令需同时在 `skills/` 创建对应目录和 `SKILL.md`
+
+* 命令文件使用 Markdown 格式，包含触发词、描述和使用示例
 
 ## Dependencies
 
 ### Internal
-- `skills/` — 每个命令对应的 skill 实现
-- `src/features/builtin-skills/` — 内置 skill 加载机制
+
+* `skills/` — 每个命令对应的 skill 实现
+
+* `src/features/builtin-skills/` — 内置 skill 加载机制
 
 <!-- MANUAL: -->

package/commands/ccg-workflow.md

@@ -11,18 +11,21 @@ skill: ccg-workflow
 ## Usage
 
 ```bash
-/ccg-workflow [new|old]
+/ccg-workflow [new | old]
 ```
 
 ## Parameters
 
-- `new` - 强制使用新项目流程
-- `old` - 强制使用老项目流程
-- 不提供参数时自动检测
+* `new` - 强制使用新项目流程
+
+* `old` - 强制使用老项目流程
+
+* 不提供参数时自动检测
 
 ## Workflow
 
 ### 新项目流程
+
 1. 需求分析 (Claude)
 2. 技术设计 (Claude)
 3. 代码开发 (Codex)
@@ -30,6 +33,7 @@ skill: ccg-workflow
 5. 测试循环 (Claude ↔ Codex)
 
 ### 老项目流程
+
 1. 读取现状 (Codex)
 2. 生成修改计划 (Claude)
 3. 模块拆分 (Codex)

package/dist/mcp/adapters/memory-adapter.d.ts

@@ -142,8 +142,8 @@ export declare const memoryTools: ({
     };
     handler: (args: {
         directive: string;
-        context?: string | undefined;
         workingDirectory?: string | undefined;
+        context?: string | undefined;
         priority?: "high" | "normal" | undefined;
     }) => Promise<{
         content: Array<{

package/dist/mcp/adapters/state-adapter.d.ts

@@ -32,8 +32,8 @@ export declare const stateTools: ({
     };
     handler: (args: {
         mode: "autopilot" | "ultrapilot" | "team" | "pipeline" | "ralph" | "ultrawork" | "ultraqa" | "swarm" | "ralplan";
-        session_id?: string | undefined;
         workingDirectory?: string | undefined;
+        session_id?: string | undefined;
     }) => Promise<{
         content: Array<{
             type: "text";
@@ -106,17 +106,17 @@ export declare const stateTools: ({
     handler: (args: {
         mode: "autopilot" | "ultrapilot" | "team" | "pipeline" | "ralph" | "ultrawork" | "ultraqa" | "swarm" | "ralplan";
         state?: Record<string, unknown> | undefined;
-        error?: string | undefined;
-        active?: boolean | undefined;
+        workingDirectory?: string | undefined;
         session_id?: string | undefined;
+        active?: boolean | undefined;
         iteration?: number | undefined;
-        workingDirectory?: string | undefined;
         max_iterations?: number | undefined;
         current_phase?: string | undefined;
         task_description?: string | undefined;
         plan_path?: string | undefined;
         started_at?: string | undefined;
         completed_at?: string | undefined;
+        error?: string | undefined;
     }) => Promise<{
         content: Array<{
             type: "text";
@@ -153,8 +153,8 @@ export declare const stateTools: ({
         required?: undefined;
     };
     handler: (args: {
-        session_id?: string | undefined;
         workingDirectory?: string | undefined;
+        session_id?: string | undefined;
     }) => Promise<{
         content: Array<{
             type: "text";
@@ -195,9 +195,9 @@ export declare const stateTools: ({
         required?: undefined;
     };
     handler: (args: {
-        session_id?: string | undefined;
         mode?: "autopilot" | "ultrapilot" | "team" | "pipeline" | "ralph" | "ultrawork" | "ultraqa" | "swarm" | "ralplan" | undefined;
         workingDirectory?: string | undefined;
+        session_id?: string | undefined;
     }) => Promise<{
         content: Array<{
             type: "text";

package/dist/mcp/adapters/trace-adapter.d.ts

@@ -25,8 +25,8 @@ export declare const traceTools: ({
     };
     handler: (args: {
         filter?: "skills" | "tools" | "all" | "hooks" | "agents" | "keywords" | "modes" | undefined;
-        sessionId?: string | undefined;
         workingDirectory?: string | undefined;
+        sessionId?: string | undefined;
         last?: number | undefined;
     }) => Promise<{
         content: Array<{
@@ -54,8 +54,8 @@ export declare const traceTools: ({
         };
     };
     handler: (args: {
-        sessionId?: string | undefined;
         workingDirectory?: string | undefined;
+        sessionId?: string | undefined;
     }) => Promise<{
         content: Array<{
             type: "text";

package/dist/mcp/config/schema.d.ts

@@ -9,8 +9,8 @@ export declare const MCPServerConfigSchema: z.ZodObject<{
     disabled: z.ZodDefault<z.ZodOptional<z.ZodBoolean>>;
     autoApprove: z.ZodDefault<z.ZodOptional<z.ZodArray<z.ZodString, "many">>>;
 }, "strip", z.ZodTypeAny, {
-    disabled: boolean;
     command: string;
+    disabled: boolean;
     autoApprove: string[];
     env?: Record<string, string> | undefined;
     args?: string[] | undefined;
@@ -30,8 +30,8 @@ export declare const MCPConfigSchema: z.ZodObject<{
         disabled: z.ZodDefault<z.ZodOptional<z.ZodBoolean>>;
         autoApprove: z.ZodDefault<z.ZodOptional<z.ZodArray<z.ZodString, "many">>>;
     }, "strip", z.ZodTypeAny, {
-        disabled: boolean;
         command: string;
+        disabled: boolean;
         autoApprove: string[];
         env?: Record<string, string> | undefined;
         args?: string[] | undefined;
@@ -44,8 +44,8 @@ export declare const MCPConfigSchema: z.ZodObject<{
     }>>;
 }, "strip", z.ZodTypeAny, {
     mcpServers: Record<string, {
-        disabled: boolean;
         command: string;
+        disabled: boolean;
         autoApprove: string[];
         env?: Record<string, string> | undefined;
         args?: string[] | undefined;

package/dist/mcp/config-schema.d.ts

@@ -9,8 +9,8 @@ export declare const MCPServerConfigSchema: z.ZodObject<{
     disabled: z.ZodDefault<z.ZodOptional<z.ZodBoolean>>;
     autoApprove: z.ZodDefault<z.ZodOptional<z.ZodArray<z.ZodString, "many">>>;
 }, "strip", z.ZodTypeAny, {
-    disabled: boolean;
     command: string;
+    disabled: boolean;
     autoApprove: string[];
     env?: Record<string, string> | undefined;
     args?: string[] | undefined;
@@ -30,8 +30,8 @@ export declare const MCPConfigSchema: z.ZodObject<{
         disabled: z.ZodDefault<z.ZodOptional<z.ZodBoolean>>;
         autoApprove: z.ZodDefault<z.ZodOptional<z.ZodArray<z.ZodString, "many">>>;
     }, "strip", z.ZodTypeAny, {
-        disabled: boolean;
         command: string;
+        disabled: boolean;
         autoApprove: string[];
         env?: Record<string, string> | undefined;
         args?: string[] | undefined;
@@ -44,8 +44,8 @@ export declare const MCPConfigSchema: z.ZodObject<{
     }>>;
 }, "strip", z.ZodTypeAny, {
     mcpServers: Record<string, {
-        disabled: boolean;
         command: string;
+        disabled: boolean;
         autoApprove: string[];
         env?: Record<string, string> | undefined;
         args?: string[] | undefined;

package/docs/AGENTS.md

@@ -9,7 +9,7 @@
 ## 关键文件
 
 | 文件 | 描述 |
-|------|------|
+| ------ | ------ |
 | `README.md` | 项目主文档 |
 | `REFERENCE.md` | 完整 API 参考（智能体、skills、hooks、工具） |
 | `ARCHITECTURE.md` | 系统架构和设计决策 |
@@ -23,7 +23,7 @@
 ## 子目录表
 
 | 目录 | 用途 | 文件数 |
-|------|------|--------|
+| ------ | ------ | -------- |
 | `standards/` | 规范体系（P0/P1 文档） | 10+ |
 | `guides/` | 用户指南和教程 | 8+ |
 | `design/` | 设计文档和决策记录 | 5+ |
@@ -44,12 +44,12 @@
    - 修改 `REFERENCE.md` 时更新智能体/skills/hooks/工具列表
    - 保持版本号和日期同步
 
-2. **编写用户指南**
+1. **编写用户指南**
    - 在 `guides/` 中创建新指南
    - 包含代码示例和验证步骤
    - 链接到相关参考文档
 
-3. **维护架构文档**
+1. **维护架构文档**
    - 修改 `ARCHITECTURE.md` 时记录设计决策
    - 更新系统图表和流程图
    - 保持与实现代码同步
@@ -57,7 +57,7 @@
 ### 修改检查清单
 
 | 修改位置 | 更新文档 |
-|---------|---------|
+| --------- | --------- |
 | 智能体数量/列表变化 | `REFERENCE.md` |
 | Skills 数量/列表变化 | `REFERENCE.md` |
 | Hooks 数量/列表变化 | `REFERENCE.md` |
@@ -70,23 +70,32 @@
 ### 常见任务
 
 ```bash
+
 # 查看参考文档
+
 cat docs/REFERENCE.md
 
 # 查看架构
+
 cat docs/ARCHITECTURE.md
 
 # 查看用户指南
+
 ls -la docs/guides/
 
 # 查看规范
+
 ls -la docs/standards/
 ```
 
 ### 文档更新规则
 
-- 保留 "Generated" 日期不变
-- 更新 "Updated" 日期为当前日期
-- 版本号与 package.json 同步
-- 所有代码示例必须经过测试
-- 链接必须指向有效的文件或 URL
+* 保留 "Generated" 日期不变
+
+* 更新 "Updated" 日期为当前日期
+
+* 版本号与 package.json 同步
+
+* 所有代码示例必须经过测试
+
+* 链接必须指向有效的文件或 URL

package/docs/ANALYTICS-SYSTEM.md

@@ -6,11 +6,15 @@ ultrapower 中 Claude API 使用情况的自动 token 追踪与费用估算。
 
 ## 功能特性
 
-- **自动 Token 追踪**：零手动记录——追踪每次 HUD 渲染
-- **费用估算**：基于模型的定价与缓存经济学
-- **会话管理**：跨多个会话追踪费用
-- **美观的 CLI**：增强的分析可视化
-- **HUD 集成**：在状态栏实时显示费用
+* **自动 Token 追踪**：零手动记录——追踪每次 HUD 渲染
+
+* **费用估算**：基于模型的定价与缓存经济学
+
+* **会话管理**：跨多个会话追踪费用
+
+* **美观的 CLI**：增强的分析可视化
+
+* **HUD 集成**：在状态栏实时显示费用
 
 ## 架构
 
@@ -36,9 +40,11 @@ StatuslineStdin → TokenExtractor → TokenTracker → Analytics Summary
 
 由于 StatuslineStdin 仅提供输入 token，输出 token 通过估算得出：
 
-- **Haiku**：输入 token 的 30%
-- **Sonnet**：输入 token 的 40%
-- **Opus**：输入 token 的 50%
+* **Haiku**：输入 token 的 30%
+
+* **Sonnet**：输入 token 的 40%
+
+* **Opus**：输入 token 的 50%
 
 所有费用显示时带有"~"前缀以表示为估算值。
 
@@ -46,52 +52,70 @@ StatuslineStdin → TokenExtractor → TokenTracker → Analytics Summary
 
 在状态栏实时显示分析数据：
 
-- 会话费用和 token 数
-- 每小时费用
-- 缓存效率
-- 预算警告（>$2 警告，>$5 严重警告）
+* 会话费用和 token 数
+
+* 每小时费用
+
+* 缓存效率
+
+* 预算警告（>$2 警告，>$5 严重警告）
 
 使用 `analytics` 预设以获得详细显示。
 
 ## CLI 用法
 
 ```bash
+
 # 查看所有内容（默认仪表盘）
+
 omc
 
 # 查看当前会话统计
+
 omc stats
 
 # 费用报告
+
 omc cost daily
 omc cost weekly
 omc cost monthly
 
 # 会话历史
+
 omc sessions
 
 # Agent 明细
+
 omc agents
 
 # 导出数据
+
 omc export cost csv ./costs.csv
 ```
 
 ## 性能
 
-- **HUD 渲染**：<100ms 总计（含分析）
-- **Token 提取**：每次渲染 <5ms
-- **摘要加载**：<10ms（mtime 缓存）
-- **CLI 启动**：<500ms
+* **HUD 渲染**：<100ms 总计（含分析）
+
+* **Token 提取**：每次渲染 <5ms
+
+* **摘要加载**：<10ms（mtime 缓存）
+
+* **CLI 启动**：<500ms
 
 ## 文件
 
-- `src/analytics/token-extractor.ts` - Token 提取
-- `src/analytics/output-estimator.ts` - 输出估算与会话 ID
-- `src/analytics/analytics-summary.ts` - 快速摘要加载
-- `src/hud/index.ts` - 自动记录集成
-- `.omc/state/token-tracking.jsonl` - 仅追加的 token 日志
-- `.omc/state/analytics-summary-{sessionId}.json` - 缓存的摘要
+* `src/analytics/token-extractor.ts` - Token 提取
+
+* `src/analytics/output-estimator.ts` - 输出估算与会话 ID
+
+* `src/analytics/analytics-summary.ts` - 快速摘要加载
+
+* `src/hud/index.ts` - 自动记录集成
+
+* `.omc/state/token-tracking.jsonl` - 仅追加的 token 日志
+
+* `.omc/state/analytics-summary-{sessionId}.json` - 缓存的摘要
 
 ## 离线转录分析
 
@@ -110,7 +134,7 @@ omc backfill [options]
 #### 选项
 
 | 选项 | 描述 |
-|--------|-------------|
+| -------- | ------------- |
 | `--project <glob>` | 过滤到特定项目路径 |
 | `--from <date>` | 仅处理在该日期之后修改的转录（ISO 格式） |
 | `--to <date>` | 仅处理在该日期之前修改的转录 |
@@ -122,16 +146,21 @@ omc backfill [options]
 #### 示例
 
 ```bash
+
 # 预览所有可用转录
+
 omc backfill --dry-run
 
 # 回填特定项目
+
 omc backfill --project "*VibeQuant*"
 
 # 仅回填最近的转录
+
 omc backfill --from "2026-01-01"
 
 # 重新处理所有内容
+
 omc backfill --reset
 ```
 
@@ -141,9 +170,11 @@ omc backfill --reset
 
 ### 数据来源
 
-- **转录位置**：`~/.claude/projects/<encoded-path>/<uuid>.jsonl`
-- **Token 数据**：来自 API 响应的实际 `output_tokens`（非估算值）
-- **Agent 检测**：从带有 Task 工具的 `tool_use` 块中提取
+* **转录位置**：`~/.claude/projects/<encoded-path>/<uuid>.jsonl`
+
+* **Token 数据**：来自 API 响应的实际 `output_tokens`（非估算值）
+
+* **Agent 检测**：从带有 Task 工具的 `tool_use` 块中提取
 
 ### 去重
 

package/docs/ARCHITECTURE.md

@@ -54,11 +54,16 @@ ultrapower transforms Claude Code into an intelligent orchestrator of specialize
 ```
 
 **Key numbers (v5.5.5):**
-- 49 specialized agents across 6 lanes (Build, Review, Domain, Product, Coordination, Axiom)
-- 71 skills for workflow automation
-- 47 hooks for event-driven lifecycle control
-- 35 custom tools (LSP x12, AST x2, REPL x1, Notepad x6, State x5, Memory x4, Trace x2, Skills x3)
-- 8 validated execution modes (autopilot, ultrapilot, team, pipeline, ralph, ultrawork, ultraqa, swarm)
+
+* 49 specialized agents across 6 lanes (Build, Review, Domain, Product, Coordination, Axiom)
+
+* 71 skills for workflow automation
+
+* 47 hooks for event-driven lifecycle control
+
+* 35 custom tools (LSP x12, AST x2, REPL x1, Notepad x6, State x5, Memory x4, Trace x2, Skills x3)
+
+* 8 validated execution modes (autopilot, ultrapilot, team, pipeline, ralph, ultrawork, ultraqa, swarm)
 
 ---
 
@@ -75,10 +80,12 @@ The Hook Bridge is the entry point for all Claude Code hook events. It normalize
 #### Normalization Flow
 
 Claude Code sends hook inputs with snake_case field names:
-- `tool_name`, `tool_input`, `tool_response`, `session_id`, `cwd`, `hook_event_name`
+
+* `tool_name`, `tool_input`, `tool_response`, `session_id`, `cwd`, `hook_event_name`
 
 `normalizeHookInput()` maps these to camelCase:
-- `toolName`, `toolInput`, `toolOutput`, `sessionId`, `directory`, `hookEventName`
+
+* `toolName`, `toolInput`, `toolOutput`, `sessionId`, `directory`, `hookEventName`
 
 A **fast-path** optimization skips Zod parsing when the input is already camelCase (detected by presence of `sessionId`, `toolName`, or `directory` keys and absence of any underscore keys). This reduces overhead for internally-generated hook calls.
 
@@ -106,7 +113,7 @@ The fast-path **does not apply** to sensitive hooks. Even if an input appears ca
 #### Hook Type Classification (15 types)
 
 | Hook Type | Sensitivity | Unknown Fields |
-|---|---|---|
+| --- | --- | --- |
 | `permission-request` | SENSITIVE (non-degradable) | Dropped |
 | `setup-init` | SENSITIVE | Dropped |
 | `setup-maintenance` | SENSITIVE | Dropped |
@@ -129,7 +136,7 @@ The Delegation Enforcer ensures that substantial code changes are routed to spec
 When a Task call specifies `subagent_type` but omits `model`, the Delegation Enforcer injects the canonical default:
 
 | Agent / Category | Injected Model |
-|---|---|
+| --- | --- |
 | `explore`, `style-reviewer`, `writer` | `haiku` |
 | `executor`, `debugger`, `verifier`, `build-fixer` | `sonnet` |
 | `designer`, `test-engineer`, `security-reviewer` | `sonnet` |
@@ -142,10 +149,13 @@ This prevents agents from defaulting to an unintended model when callers omit th
 
 The enforcer applies these rules to determine whether direct handling or delegation is required:
 
-- **Multi-file implementation** → must delegate to `executor` (or `deep-executor` for complex autonomous tasks)
-- **Complex refactoring** → must delegate to `deep-executor`
-- **Architecture decisions** → must delegate to `architect`
-- **Single-line or trivial changes** → may be handled directly
+* **Multi-file implementation** → must delegate to `executor` (or `deep-executor` for complex autonomous tasks)
+
+* **Complex refactoring** → must delegate to `deep-executor`
+
+* **Architecture decisions** → must delegate to `architect`
+
+* **Single-line or trivial changes** → may be handled directly
 
 Violations emit a warning to stderr; the enforcer does not hard-block (to avoid interrupting valid edge cases).
 
@@ -194,7 +204,7 @@ Note (D-03): The original PRD listed 7 modes; `swarm` was added in a subsequent
 #### State File Paths
 
 | Scope | Path |
-|---|---|
+| --- | --- |
 | Global mode state | `{worktree}/.omc/state/{mode}-state.json` |
 | Session-scoped state | `{worktree}/.omc/state/sessions/{sessionId}/{mode}-state.json` |
 | Swarm (SQLite) | `{worktree}/.omc/state/swarm-state.db` |
@@ -208,7 +218,7 @@ All state access must go through the `state_read` / `state_write` MCP tools, not
 The Feature Layer sits above the Core Pipeline and provides cross-cutting capabilities consumed by skills, agents, and hooks.
 
 | Module | Location | Purpose |
-|---|---|---|
+| --- | --- | --- |
 | `model-routing` | `src/features/model-routing/` | Task complexity → model tier mapping (Haiku/Sonnet/Opus) |
 | `boulder-state` | `src/features/boulder-state/` | Ralph/ultrawork "boulder never stops" persistence |
 | `verification` | `src/features/verification/` | Evidence-based completion checks (BUILD/TEST/LINT/TODO) |
@@ -263,7 +273,7 @@ State transitions are persisted in `.omc/axiom/active_context.md`. The system re
 Three mandatory gates prevent low-quality work from advancing:
 
 | Gate | Trigger | Condition |
-|---|---|---|
+| --- | --- | --- |
 | **Expert Gate** | All new feature requirements | Must pass `/ax-draft` → `/ax-review` (5 parallel expert reviews) |
 | **User Gate** | PRD finalized | Display "PRD generated, confirm? (Yes/No)" — blocks until confirmed |
 | **CI Gate** | Code changes complete | Must run `tsc --noEmit && npm run build && npm test` with zero errors |
@@ -289,7 +299,7 @@ Axiom uses a layered file-based memory system under `.omc/axiom/`:
 ### Axiom Agents (14)
 
 | Agent | Purpose |
-|---|---|
+| --- | --- |
 | `axiom-requirement-analyst` | Three-gate requirement screening (PASS/CLARIFY/REJECT) |
 | `axiom-product-designer` | Draft PRD generation with Mermaid flow diagrams |
 | `axiom-review-aggregator` | 5-expert parallel review aggregation and conflict arbitration |
@@ -308,7 +318,7 @@ Axiom uses a layered file-based memory system under `.omc/axiom/`:
 ### Axiom Skills (14)
 
 | Skill | Command | Purpose |
-|---|---|---|
+| --- | --- | --- |
 | `ax-draft` | `/ax-draft` | Requirement clarification → Draft PRD → user confirm |
 | `ax-review` | `/ax-review` | 5-expert parallel review → aggregate → Rough PRD |
 | `ax-decompose` | `/ax-decompose` | Rough PRD → system architecture → atomic task DAG |
@@ -327,7 +337,7 @@ Axiom uses a layered file-based memory system under `.omc/axiom/`:
 ### Axiom Hooks (2)
 
 | Hook | Location | Purpose |
-|---|---|---|
+| --- | --- | --- |
 | `axiom-boot` | `src/hooks/axiom-boot/` | Session startup memory context injection |
 | `axiom-guards` | `src/hooks/axiom-guards/` | Gate rule enforcement (Expert/User/CI Gate) |
 
@@ -338,7 +348,7 @@ Axiom uses a layered file-based memory system under `.omc/axiom/`:
 ultrapower ships four pre-packaged MCP servers under `bridge/`:
 
 | Server | Model | Primary Use |
-|---|---|---|
+| --- | --- | --- |
 | `codex-server` | gpt-5.3-codex | Architecture review, planning validation, code analysis |
 | `gemini-server` | gemini-3-pro-preview | Large-context tasks (1M tokens), UI/UX design review |
 | `mcp-server` | — | General MCP tool service (tool discovery, skill loading) |
@@ -353,7 +363,7 @@ If discovery returns no results, the MCP server is not configured. The system fa
 ### Provider Selection Guide
 
 | Task Type | Preferred Provider | Agent Role |
-|---|---|---|
+| --- | --- | --- |
 | Architecture review, planning validation | Codex | `architect`, `planner`, `critic` |
 | Security review, code review | Codex | `security-reviewer`, `code-reviewer` |
 | Test strategy | Codex | `test-engineer` |
@@ -447,7 +457,7 @@ HIGH   → Opus    (complex reasoning, architecture, deep debugging)
 ### Agent Tier Assignments
 
 | Lane | Agent | Default Model |
-|---|---|---|
+| --- | --- | --- |
 | Build/Analysis | `explore` | haiku |
 | Build/Analysis | `analyst`, `planner`, `architect` | opus |
 | Build/Analysis | `debugger`, `executor`, `verifier` | sonnet |
@@ -493,7 +503,7 @@ Skills are behavior injectors that modify how the orchestrator runs. They compos
 ### Execution Modes
 
 | Mode | Trigger Keywords | Purpose |
-|---|---|---|
+| --- | --- | --- |
 | `autopilot` | "autopilot", "build me", "I want a" | Fully autonomous idea-to-code |
 | `ultrawork` | "ulw", "ultrawork" | Maximum parallelism agent orchestration |
 | `ralph` | "ralph", "don't stop" | Persistent loop until verifier confirms done |
@@ -512,7 +522,7 @@ Hooks provide event-driven lifecycle control across 15 hook event types register
 ### Key Hook Modules
 
 | Hook Module | Location | Event | Purpose |
-|---|---|---|---|
+| --- | --- | --- | --- |
 | `keyword-detector` | `src/hooks/keyword-detector/` | UserPromptSubmit | Magic keyword detection → skill/mode activation |
 | `autopilot` | `src/hooks/autopilot/` | UserPromptSubmit | Autopilot mode loop management |
 | `ralph` | `src/hooks/ralph/` | SubagentStop | Ralph persistence loop |
@@ -579,7 +589,7 @@ state_get_status(mode?)             // Detailed status for one or all modes
 Before claiming completion, all work must pass the verification protocol:
 
 | Check | What is Verified |
-|---|---|
+| --- | --- |
 | BUILD | TypeScript compiles without errors (`tsc --noEmit && npm run build`) |
 | TEST | All tests pass (`npm test`) |
 | LINT | No lint errors (`npm run lint`) |
@@ -590,16 +600,19 @@ Before claiming completion, all work must pass the verification protocol:
 Evidence must be fresh (within 5 minutes) and include actual command output. The `verifier` agent collects evidence and reports with explicit pass/fail per check.
 
 **Model selection for verification:**
-- Small changes (<5 files, <100 lines): `verifier` with `model="haiku"`
-- Standard changes: `verifier` with `model="sonnet"`
-- Large or security/architecture changes (>20 files): `verifier` with `model="opus"`
+
+* Small changes (<5 files, <100 lines): `verifier` with `model="haiku"`
+
+* Standard changes: `verifier` with `model="sonnet"`
+
+* Large or security/architecture changes (>20 files): `verifier` with `model="opus"`
 
 ---
 
 ## Further Reading
 
 | Document | Content |
-|---|---|
+| --- | --- |
 | [REFERENCE.md](./REFERENCE.md) | Complete agent, skill, hook, and tool reference |
 | [FEATURES.md](./FEATURES.md) | Internal API and feature module documentation |
 | [MIGRATION.md](./MIGRATION.md) | Breaking changes and migration guides |