feat(agent): replace workflow modes with built-in skills

CONTRIBUTING.md

@@ -118,7 +118,7 @@ We welcome contributions beyond standard feature or bug-fix PRs. Examples includ
 | Prompts | `src/crates/assembly/core/src/agentic/agents/prompts/` | Add or refine prompts, and update related logic as needed |
 | Tools | `src/crates/assembly/core/src/agentic/tools/implementations/`, `src/crates/assembly/core/src/agentic/tools/registry.rs` | Add tool implementations and register them in the tool registry |
 | Subagents | `src/crates/assembly/core/src/agentic/agents/custom_subagents/`, `src/crates/assembly/core/src/agentic/agents/registry.rs` | Add subagent implementations and register them in the subagent registry |
-| Mode contributions | `src/crates/assembly/core/src/agentic/agents/*_mode.rs`, `src/crates/assembly/core/src/agentic/agents/prompts/*_mode.md`, `src/web-ui/src/locales/*/settings/modes.json` | Add/improve agent modes (e.g. Plan/Debug/Agentic or custom modes) and keep prompts + UI copy in sync |
+| Mode contributions | `src/crates/assembly/core/src/agentic/agents/*_mode.rs`, `src/crates/assembly/core/src/agentic/agents/prompts/*_mode.md`, `src/web-ui/src/locales/*/settings/modes.json` | Add/improve Agentic or custom modes and keep prompts + UI copy in sync |
 | Scenario guides for Code Agent and AIIde | `website/src/docs/` | Add workflows, playbooks, and real-world scenario docs (or link them from `README.md`) |
 
 ### Before you start

CONTRIBUTING_CN.md

@@ -108,7 +108,7 @@ DevTools；`Cmd/Ctrl + Shift + I` 切换 BitFun 元素检查器，`Cmd/Ctrl + Sh
 | Prompts | `src/crates/assembly/core/src/agentic/agents/prompts/` | 新增或优化提示词，并按需更新相关逻辑 |
 | Tools | `src/crates/assembly/core/src/agentic/tools/implementations/`、`src/crates/assembly/core/src/agentic/tools/registry.rs` | 新增工具实现，并在工具注册表中注册 |
 | Subagents | `src/crates/assembly/core/src/agentic/agents/custom_subagents/`、`src/crates/assembly/core/src/agentic/agents/registry.rs` | 新增子代理实现，并在子代理注册表中注册 |
-| 模式贡献 | `src/crates/assembly/core/src/agentic/agents/*_mode.rs`、`src/crates/assembly/core/src/agentic/agents/prompts/*_mode.md`、`src/web-ui/src/locales/*/settings/modes.json` | 新增/优化 Agent 模式（例如 Plan/Debug/Agentic 或自定义模式）的逻辑与提示词，并同步前端模式文案 |
+| 模式贡献 | `src/crates/assembly/core/src/agentic/agents/*_mode.rs`、`src/crates/assembly/core/src/agentic/agents/prompts/*_mode.md`、`src/web-ui/src/locales/*/settings/modes.json` | 新增/优化 Agentic 或自定义 Agent 模式的逻辑与提示词，并同步前端模式文案 |
 | Code Agent 与 AIIde 场景指南 | `website/src/docs/` | 补充流程、playbook 与真实场景说明（或从 `README.md` 链接） |
 
 ### 开始前

README.md

@@ -35,7 +35,7 @@ Writes code, produces documents, and drives the desktop — with Mini Apps, a Ru
 | --- | --- |
 | **Agentic Mini Apps** | A task gets its own interface — chart, board, form, panel — with a conversation bound to that interface's live state |
 | **Self-hosted multi-device control** | Login, cross-device session sync, and controlling one device from another run through a relay you deploy. Zero-knowledge; no vendor cloud in the path |
-| **Coding** | Plan, edit, test, and commit inside real Git repositories. Agentic, Plan, Debug, Deep Review, long-horizon tasks |
+| **Coding** | Plan, edit, test, and commit inside real Git repositories. Agentic, planning and debugging Skills, Deep Review, long-horizon tasks |
 | **Office work** | Research, writing, presentations, meeting notes, reports |
 | **Desktop execution** | Browser, terminal, desktop applications, the filesystem, and remote workspaces |
 | **Four tiers of customization** | Custom Agents → MCP / Skills / Hooks → Mini Apps → source-level changes |
@@ -90,7 +90,7 @@ Two kinds of complex work: shipping code in real repositories, and turning sourc
 
 | Scenario | Delivery goal | Typical capabilities |
 | --- | --- | --- |
-| **Coding** | Move from a real repository to a mergeable result. | Agentic, Plan, Debug, testing, Git, Deep Review, long-horizon tasks, and benchmarks. |
+| **Coding** | Move from a real repository to a mergeable result. | Agentic, planning and debugging Skills, testing, Git, Deep Review, long-horizon tasks, and benchmarks. |
 | **Office Work** | Move from source material to useful written and visual deliverables. | Research, presentations, summarization, writing, meeting notes, and reports. |
 
 **Shared capabilities**

README.zh-CN.md

@@ -35,7 +35,7 @@
 | --- | --- |
 | **Agentic Mini App** | 为任务生成专属界面——图表、看板、表单、面板——对话绑定该界面的实时状态 |
 | **自部署多设备互联互控** | 账号登录、跨设备会话同步、设备间操控，全部走你自己部署的 relay。零知识加密，不经第三方云 |
-| **编码交付** | 在真实 Git 仓库里规划、改代码、跑测试、提交。Agentic、Plan、Debug、Deep Review、长程任务 |
+| **编码交付** | 在真实 Git 仓库里规划、改代码、跑测试、提交。Agentic、规划与调试 Skill、Deep Review、长程任务 |
 | **办公交付** | 调研、写作、演示文稿、会议纪要、报告 |
 | **桌面执行层** | 浏览器、终端、桌面软件、文件系统、远程工作区 |
 | **四层可定制** | 自定义 Agent → MCP / Skills / Hooks → Mini App → 源码级改造 |
@@ -90,7 +90,7 @@ pnpm run desktop:dev
 
 | 场景 | 目标交付 | 典型能力 |
 | --- | --- | --- |
-| **编码** | 从真实仓库推进到可合并结果。 | Agentic、Plan、Debug、测试、Git、Deep Review、长程任务、Benchmark。 |
+| **编码** | 从真实仓库推进到可合并结果。 | Agentic、规划与调试 Skill、测试、Git、Deep Review、长程任务、Benchmark。 |
 | **办公** | 从资料推进到实用的文字和视觉交付物。 | 调研、演示文稿、总结、写作、会议纪要、报告。 |
 
 **通用能力**

docs/architecture/model-request-cache-reuse.md

@@ -176,49 +176,24 @@ and DeepReview retry fields.
 That restriction is not just validation polish; it protects cache reuse by
 keeping the forked child aligned with the parent session shape.
 
-## 3. The Shared Coding Modes Intentionally Reuse The Same Cache Identity
+## 3. Coding Workflows Keep The Agentic Cache Identity Stable
 
-BitFun's four shared coding modes are:
-
-- `agentic`
-- `Plan`
-- `debug`
-- `Multitask`
-
-They are intentionally configured to share the same stable prompt base.
+`agentic` owns the stable coding prompt base. Planning and parallel task
+coordination are built-in `plan` and `multitask` Skills rather than separate
+Agent modes.
 
 Relevant code:
 
-- shared constants and tests:
+- the Agentic prompt identity and coding profile facts:
   `src/crates/execution/agent-runtime/src/agents.rs`
-- mode definitions:
-  `src/crates/assembly/core/src/agentic/agents/definitions/modes/{agentic,plan,debug,multitask}.rs`
-
-Why they reuse cache:
-
-- all four modes use the same prompt template:
-  `SHARED_CODING_MODE_PROMPT_TEMPLATE = "agentic_mode"`
-- all four modes use the same user-context policy:
-  `shared_coding_mode_user_context_policy()`
-- `Agent::system_prompt_cache_identity(...)` is derived from
-  `prompt_template_name(...)`
-- `Agent::user_context_cache_identity(...)` is derived from the user-context
-  policy scope key
-
-The test `shared_template_modes_share_system_prompt_cache_identity()` asserts
-that these shared modes intentionally produce the same system-prompt and
-user-context cache identities.
-
-Mode-specific behavior is added through `system_reminder` text, not by swapping
-the cached base template:
-
-- `Plan`, `Debug`, and `Multitask` provide first-entry and ongoing reminder
-  templates
-- reminders are injected immediately before the current user message in
-  `ExecutionEngine::build_ai_messages_for_send(...)`
-
-This is the key reason mode switches between these four coding modes do not
-force a base prompt cache reset.
+- the Agentic definition:
+  `src/crates/assembly/core/src/agentic/agents/definitions/modes/agentic.rs`
+- built-in Skill catalog and mode policy:
+  `src/crates/execution/agent-runtime/src/skills/`
+
+Skill availability and loading use the dynamic listing path described below.
+Invoking a coding workflow therefore does not create another base Agent prompt
+identity; only the dynamic Skill context changes.
 
 The frontend also knows about this compatibility:
 

docs/interactive-capabilities/capabilities.json

@@ -4,7 +4,7 @@
   "title": "BitFun Playbook",
   "origin": "https://playbook.openbitfun.com",
   "source": "src/shared/interactive-capabilities/catalog.json",
-  "digest": "b0794e2a65d5356a8fdb262f0acf26753ac1aebfedf9b1e252d4dce4a21bc9a1",
+  "digest": "5a170c4fb33ea5bb4fe1d839ff0d405db53b82e586aead0891babafbb59b40aa",
   "ownerDigest": "a6a8c5337fb877e641b66f5774740d5213caed4800fe9b006eb53632f7eb1c2d",
   "searchAcceptance": [
     {
@@ -796,7 +796,6 @@
       },
       "delegateTools": [
         "TodoWrite",
-        "CreatePlan",
         "AskUserQuestion"
       ],
       "presentationTarget": {
@@ -18525,15 +18524,14 @@
             "kind": "delegate",
             "tools": [
               "TodoWrite",
-              "CreatePlan",
               "AskUserQuestion"
             ],
             "workflowZh": [
-              "复杂任务用 TodoWrite 维护当前进度，只有需要正式计划产物时才用 CreatePlan",
+              "复杂任务用 TodoWrite 维护当前进度；正式计划产物通过 Write 写入项目的 .bitfun/plans 目录",
               "缺少会改变结果的用户选择时用 AskUserQuestion；能从仓库或当前上下文查明的信息不重复询问"
             ],
             "workflowEn": [
-              "Use TodoWrite to maintain progress for complex work and CreatePlan only when a formal plan artifact is needed",
+              "Use TodoWrite to maintain progress for complex work; create formal plan artifacts with Write under the project's .bitfun/plans directory",
               "Use AskUserQuestion only for a missing choice that changes the outcome; do not ask again for facts available from the repository or current context"
             ]
           }
@@ -18703,11 +18701,10 @@
         "Inspect the current goal with get_goal; create one with create_goal only when the user explicitly asks",
         "Use update_goal complete only after the objective is achieved, and blocked only after the repeated-blocker threshold is genuinely met",
         "TodoWrite",
-        "CreatePlan",
         "AskUserQuestion",
-        "复杂任务用 TodoWrite 维护当前进度，只有需要正式计划产物时才用 CreatePlan",
+        "复杂任务用 TodoWrite 维护当前进度；正式计划产物通过 Write 写入项目的 .bitfun/plans 目录",
         "缺少会改变结果的用户选择时用 AskUserQuestion；能从仓库或当前上下文查明的信息不重复询问",
-        "Use TodoWrite to maintain progress for complex work and CreatePlan only when a formal plan artifact is needed",
+        "Use TodoWrite to maintain progress for complex work; create formal plan artifacts with Write under the project's .bitfun/plans directory",
         "Use AskUserQuestion only for a missing choice that changes the outcome; do not ask again for facts available from the repository or current context"
       ],
       "docsUrl": "https://playbook.openbitfun.com/capabilities/feature.ai-assistant/"

docs/interactive-capabilities/capabilities/feature.ai-assistant.md

@@ -39,7 +39,7 @@ Work with AI in project context, create sessions, continue tasks, answer permiss
   - Inspect, create, complete, or block the current session goal
 - **Agent 可定位入口，需交互完成 / Agent opens; interaction required** · 编辑目标说明或清除当前会话目标
   - Edit the goal objective or clear the current session goal
-- **由专用 Agent 工具控制 / Delegated Agent tool** · `TodoWrite` / `CreatePlan` / `AskUserQuestion` · 让 Agent 维护任务清单、生成实施计划并按需向用户提问
+- **由专用 Agent 工具控制 / Delegated Agent tool** · `TodoWrite` / `AskUserQuestion` · 让 Agent 维护任务清单、生成实施计划并按需向用户提问
   - Let an agent maintain a task list, produce an implementation plan, and ask the user when needed
 - **Agent 可定位入口，需交互完成 / Agent opens; interaction required** · 查看单个会话的请求、Token 与成本用量报告
   - View request, token, and cost usage for an individual session

docs/interactive-capabilities/technical/product-control-open-audit.json

@@ -1,7 +1,7 @@
 {
   "schemaVersion": 1,
   "generatedFrom": "src/shared/interactive-capabilities/catalog.json",
-  "catalogDigest": "b0794e2a65d5356a8fdb262f0acf26753ac1aebfedf9b1e252d4dce4a21bc9a1",
+  "catalogDigest": "5a170c4fb33ea5bb4fe1d839ff0d405db53b82e586aead0891babafbb59b40aa",
   "count": 207,
   "reasonCounts": {
     "externalAuth": 4,

docs/interactive-capabilities/technical/tauri-command-map.json

@@ -1,7 +1,7 @@
 {
   "schemaVersion": 2,
   "generatedFrom": "src/shared/interactive-capabilities/catalog.json",
-  "catalogDigest": "b0794e2a65d5356a8fdb262f0acf26753ac1aebfedf9b1e252d4dce4a21bc9a1",
+  "catalogDigest": "5a170c4fb33ea5bb4fe1d839ff0d405db53b82e586aead0891babafbb59b40aa",
   "commandCount": 651,
   "coverage": {
     "commandCount": 651,

docs/sdlc-harness/features/agent-evaluation.md

@@ -40,7 +40,7 @@
 
 范围：
 
-- 评测 Plan、Debug、Review、严格审查、Agentic、Mini App、remote、MCP 等模式。
+- 评测 Plan/Debug Skill、Review、严格审查、Agentic、Mini App、remote、MCP 等工作流与能力。
 - 评测工具接口、上下文策略、hook 策略、审批/安全策略。
 - 评测配置化策略是否误升级、漏提示或过度打断。
 - 评测安全边界是否降低风险且不制造无意义确认。

docs/sdlc-harness/product-requirements-agent-workflow-adjustment.md

@@ -63,7 +63,7 @@ BitFun 后续不应把 dynamic workflow 理解成一个需要用户学习的新
 
 ### 4.1 从固定模式改为自适应执行
 
-现有体验容易把 Agentic、Plan、Debug、Review、DeepReview 等理解为不同模式。后续应调整为：
+现有体验曾经容易把 Agentic、Plan、Debug、Review、DeepReview 等理解为不同模式；Plan 和 Debug 现已收敛为 Skill，后续仍应继续按以下方向调整：
 
 ```text
 用户表达目标

scripts/core-boundaries/rules/source/required-rules.mjs

@@ -7897,21 +7897,6 @@ export const requiredContentRules = [
       },
     ],
   },
-  {
-    path: 'src/crates/assembly/core/src/agentic/agents/definitions/modes/mod.rs',
-    reason:
-      'core agent mode definitions must continue exposing Multitask mode until an approved agent-runtime migration preserves mode registration semantics',
-    patterns: [
-      {
-        regex: /\bmod multitask\b/,
-        message: 'missing Multitask mode module',
-      },
-      {
-        regex: /\bpub use multitask::MultitaskMode\b/,
-        message: 'missing Multitask mode export',
-      },
-    ],
-  },
   {
     path: 'src/crates/assembly/core/src/agentic/agents/definitions/subagents/mod.rs',
     reason:
@@ -7951,10 +7936,6 @@ export const requiredContentRules = [
         regex: /\bbuiltin_agent_definition_specs\(\)/,
         message: 'missing builtin agent definition catalog owner',
       },
-      {
-        regex: /builtin_agent_spec\(\s*"Multitask",\s*Mode,\s*"auto"/,
-        message: 'missing Multitask runtime default model mapping',
-      },
       {
         regex: /builtin_agent_spec\(\s*"GeneralPurpose",\s*SubAgent,\s*"primary"/,
         message: 'missing GeneralPurpose runtime default model mapping',

scripts/core-boundaries/self-test.mjs

@@ -4204,10 +4204,6 @@ export function runManifestParserSelfTest({
       path: 'src/crates/assembly/core/src/agentic/agents/registry/types.rs',
       contracts: ['SubagentQueryContext', 'SubagentListScope', 'default_enabled', 'effective_enabled', 'SubagentStateReason'],
     },
-    {
-      path: 'src/crates/assembly/core/src/agentic/agents/definitions/modes/mod.rs',
-      contracts: ['mod multitask', 'MultitaskMode'],
-    },
     {
       path: 'src/crates/assembly/core/src/agentic/agents/definitions/subagents/mod.rs',
       contracts: ['mod general_purpose', 'GeneralPurposeAgent'],

scripts/core-boundaries/source-content-checks.mjs

@@ -1,5 +1,5 @@
 import { spawnSync } from 'node:child_process';
-import { readFileSync } from 'node:fs';
+import { existsSync, readFileSync } from 'node:fs';
 import { join } from 'node:path';
 
 export function listTrackedRustRepoPaths(root) {
@@ -11,7 +11,13 @@ export function listTrackedRustRepoPaths(root) {
   if (result.status !== 0) {
     throw new Error(result.stderr.trim() || 'git ls-files failed');
   }
-  return result.stdout.split('\0').filter(Boolean);
+  return result.stdout
+    .split('\0')
+    .filter(Boolean)
+    // `git ls-files` continues to report tracked files deleted only in the
+    // working tree. Boundary scans must evaluate the prospective checkout
+    // without requiring callers to stage deletions first.
+    .filter((repoPath) => existsSync(join(root, ...repoPath.split('/'))));
 }
 
 function stripRustComments(text) {

scripts/i18n-contract.test.mjs

@@ -271,11 +271,11 @@ test('web-ui slash command picker distinguishes all commands from quick actions'
   const source = readText('src/web-ui/src/flow_chat/components/ChatInput.tsx');
   const allCommandsStart = source.indexOf("if (slashCommandState.kind === 'all')");
   const allCommandsEnd = source.indexOf(
-    'if (!canSwitchModes || !showStandardExecutionOptions)',
+    "if (slashCommandState.kind === 'skills')",
     allCommandsStart,
   );
   assert.notEqual(allCommandsStart, -1, 'ChatInput should render an all-command slash picker state');
-  assert.notEqual(allCommandsEnd, -1, 'ChatInput all-command branch should stay before the mode-only branch');
+  assert.notEqual(allCommandsEnd, -1, 'ChatInput all-command branch should stay before the skills-only branch');
 
   const allCommandsBlock = source.slice(allCommandsStart, allCommandsEnd);
   assert.match(

src/apps/cli/src/main.rs

@@ -2199,8 +2199,8 @@ mod harness_profile_compatibility_tests {
             "minimal"
         );
         assert_eq!(
-            agent_type_with_harness_profile("Plan".to_string(), Some("balanced")),
-            "Plan"
+            agent_type_with_harness_profile("Cowork".to_string(), Some("balanced")),
+            "Cowork"
         );
     }
 

src/apps/cli/src/modes/chat.rs

@@ -488,7 +488,7 @@ const LINEAGE_SETTLEMENT_RETRY_MAX: Duration = Duration::from_secs(1);
 pub(crate) struct ChatMode {
     config: CliConfig,
     keymap: ResolvedKeymap,
-    /// Current agent type (e.g. "agentic", "Plan", "Multitask")
+    /// Current agent type (e.g. "agentic", "Cowork", "DeepResearch")
     agent_type: String,
     workspace: Option<String>,
     local_cwd: std::path::PathBuf,

src/apps/cli/src/modes/chat/tests.rs

@@ -1818,7 +1818,7 @@ mod tests {
     fn previous_session_update_failure_is_not_reported_as_a_success() {
         let status = previous_session_update_status(
             "mode",
-            "Plan",
+            "Cowork",
             &SessionUpdateApplyOutcome::SessionUpdateFailed("storage unavailable".to_string()),
         );
 

src/apps/cli/tests/acp_stdio_cli.rs

@@ -215,14 +215,14 @@ async fn acp_stdio_preserves_mode_and_history_across_restart_then_closes_active_
             json!({
                 "sessionId": session_id,
                 "configId": "mode",
-                "value": " Plan "
+                "value": " Cowork "
             }),
         )
         .await;
     assert!(configured.get("error").is_none(), "{configured}");
     assert_eq!(
         current_config_value(&configured, "mode"),
-        Some(&json!("Plan"))
+        Some(&json!("Cowork"))
     );
     assert!(
         mode_updates.iter().all(|message| {
@@ -335,7 +335,7 @@ async fn acp_stdio_preserves_mode_and_history_across_restart_then_closes_active_
         .find(|response| response.get("error").is_some())
         .expect("one concurrent session/load must be rejected");
     assert!(loaded.get("error").is_none(), "{loaded}");
-    assert_eq!(current_config_value(loaded, "mode"), Some(&json!("Plan")));
+    assert_eq!(current_config_value(loaded, "mode"), Some(&json!("Cowork")));
     assert!(
         !replay_updates.is_empty(),
         "session/load must replay persisted history before its success response"
@@ -383,7 +383,7 @@ async fn acp_stdio_preserves_mode_and_history_across_restart_then_closes_active_
     );
     assert_eq!(
         current_config_value(&reloaded_after_close, "mode"),
-        Some(&json!("Plan"))
+        Some(&json!("Cowork"))
     );
     assert_eq!(
         replay_after_close