@vitos-pizza/vitos-pizza 0.2.0 → 0.3.1

package/AGENTS.md

@@ -18,7 +18,7 @@ Declare npm pi packages in root `package.json`:
 - `bundledDependencies` — include in the distribution tarball
 - `piBundled` — list read by `sync-pi-manifest.mjs` to merge `node_modules/<pkg>/` paths into the root `pi` manifest
 
-Extension load order: topological sort of `pi.requires`, with `LOCAL_ORDER` tie-break. Default: `permission-system` → `question` → `ui-enhancements` → `agent-mode` → `subagents` → `session-title` → `keybindings`.
+Extension load order: topological sort of `pi.requires`, with `LOCAL_ORDER` tie-break. Default: `permission-system` → `settings-preset` → `question` → `ui-enhancements` → `subagents` → `websearch` → `session-title` → `keybindings` → `agent-mode` → `todoist`. (`agent-mode` requires `keybindings` so shortcut actions register after the keybindings listener is ready.)
 
 ## Agent mode
 
@@ -30,7 +30,7 @@ Vito's Pizzeria uses three centralized modes via **`@vitos-pizza/agent-mode`**:
 | `plan` | `plan` preset | Read-only; scout → planner via subagents + `question` |
 | `execute` | `yolo` preset | Minimal gates; full implementation |
 
-Switch with `/mode [agent|plan|execute]` or **`F6`** (`agent-mode.cycle`). Current mode appears in the ui-enhancements border bar (`· plan` / `· execute`).
+Switch with `/mode [agent|plan|execute]` or **`Ctrl+.`** / **`Alt+M`** (`agent-mode.cycle`). Current mode appears in the ui-enhancements border bar (`· plan` / `· execute`).
 
 ## Keyboard shortcuts
 
@@ -44,11 +44,13 @@ import { emitShortcutAction } from "@vitos-pizza/keybindings/types";
 // session_start — register handler only; keys come from vitos-shortcuts.json
 emitShortcutAction(pi.events, {
   id: "agent-mode.cycle",
-  description: "Cycle agent mode (agent → plan → execute)",
-  handler: async (ctx) => { /* ... */ },
+ description: "Cycle agent mode (agent → plan → execute)",
+ handler: async (ctx) => { /* ... */ },
 });
 ```
 
+Default binding: `ctrl+.` and `alt+m` (see `packages/keybindings/presets/shortcuts.json`). `Ctrl+.` needs a terminal with Kitty / modifyOtherKeys; `Alt+M` works in classic terminals.
+
 Key config (project overrides global overrides preset defaults):
 
 - `packages/keybindings/presets/shortcuts.json` — distribution defaults

package/CHANGELOG.md

@@ -2,6 +2,33 @@
 
 All notable changes to vitos-pizza are documented here.
 
+## [0.3.1] - 2026-07-11
+
+### Changed
+
+- **Prompt optimization** — plan mode: authoritative short system instructions + supersede; sticky reminder without JSON dumps
+- Subagents skill: mode constraints one-liners; plan vs execute/agent examples split (no worker chain in plan)
+- Scout / planner / worker / title prompts tightened; planner output adds `## 功能验收`
+- Hide internal `title` agent from public subagent listings
+- `web_search` guidelines: prefer local tools; use web only for current external facts
+
+### Acknowledgments
+
+- Claude Code plan-mode reminder pattern and OpenAI Codex base-instruction style (prompt craft reference)
+
+## [0.3.0] - 2026-07-11
+
+### Added
+
+- **Execute-mode prompt UI** — bordered editor shows `▶ 开始实现` + free input (status bar preserved)
+- **`@vitos-pizza/settings-preset`** — seeds default Pi settings on first session when none exist
+- Mode cycle bindings: **`Ctrl+.`** and **`Alt+M`** (`agent-mode.cycle`)
+
+### Changed
+
+- Shortcut registration: actions register on `session_start` so keybindings can bind them reliably
+- Plan mode: main agent / scout / planner can use `question` when helpful; scout tools aligned to plan-friendly set (no bash/write)
+
 ## [0.2.0] - 2026-07-11
 
 ### Added

package/README.md

@@ -25,10 +25,10 @@ npm install -g --ignore-scripts @earendil-works/pi-coding-agent
 
 ```bash
 # npm (recommended)
-pi install npm:@vitos-pizza/vitos-pizza@0.2.0
+pi install npm:@vitos-pizza/vitos-pizza@0.3.1
 
 # git (pinned tag)
-pi install git:github.com/xin2017338/vitos-pizza@v0.2.0
+pi install git:github.com/xin2017338/vitos-pizza@v0.3.1
 ```
 
 From source:
@@ -117,7 +117,7 @@ subagent({ tasks: [{ agent: "scout", task: "scan src" }, { agent: "scout", task:
 subagent({ agent: "worker", task: "...", async: true })  →  wait({ id: "<runId>" })
 ```
 
-Built-in agents: **scout**, **planner**, **worker**, **title** (session auto-naming). Other modules request runs over `pi.events` RPC — import `requestSubagentRun` from `@vitos-pizza/subagents/rpc/client`.
+Built-in agents: **scout**, **planner**, **worker**. Other modules request runs over `pi.events` RPC — import `requestSubagentRun` from `@vitos-pizza/subagents/rpc/client`.
 
 ## Web search
 
@@ -148,7 +148,7 @@ Built-in **@vitos-pizza/permission-system** enforces allow / ask / deny on tools
 /mode plan                    # or agent | execute
 ```
 
-Or press **`F6`** to cycle agent → plan → execute. Use `/vitos-shortcuts` to view bindings.
+Or press **`Ctrl+.`** or **`Alt+M`** to cycle agent → plan → execute. Use `/vitos-shortcuts` to view bindings.
 
 | Mode | Permissions | Behavior |
 |------|-------------|----------|
@@ -216,6 +216,8 @@ vitos-pizza builds on the Pi ecosystem and several open-source projects. We are
 | [Pi `border-status-editor` example](https://github.com/earendil-works/pi/tree/main/packages/coding-agent/examples/extensions/border-status-editor.ts) | `@vitos-pizza/ui-enhancements` | Editor-border status bar (model, thinking, context, cwd, git) |
 | [sf-welcome](https://github.com/salesforce/sf-pi/blob/main/extensions/sf-welcome/index.ts) | `@vitos-pizza/ui-enhancements` | Welcome header session lifecycle and tips layout |
 | **Claude Code** permission UX | Agent modes (`agent` / `plan` / `execute`) | Preset naming and mode-switching inspiration (not a code dependency) |
+| [Claude Code system prompts](https://github.com/CANYOUFINDIT/claude-code-system-prompts) (plan-mode reminders) | `@vitos-pizza/agent-mode` plan instructions | Short sticky reminder + supersede pattern for plan mode (not a code dependency) |
+| [OpenAI Codex base instructions](https://github.com/openai/codex/blob/main/codex-rs/protocol/src/prompts/base_instructions/default.md) | Plan / subagent prompt style | Concise hard rules and precedence over conflicting guidance (not a code dependency) |
 | [Claude Code TodoWrite / Task tools](https://code.claude.com/docs/en/agent-sdk/todo-tracking.md) | `@vitos-pizza/todoist` | Complete = status done; delete only when no longer relevant (cancelled / mistaken) |
 | [@capyup/pi-basic-tools todo](https://www.npmjs.com/package/@capyup/pi-basic-tools) | `@vitos-pizza/todoist` | completed vs deleted prompt boundary for task-list tools |
 | [pi-package-template](https://github.com/S1M0N38/pi-package-template) | Repo layout | Pi package monorepo conventions |

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@vitos-pizza/vitos-pizza",
-	"version": "0.2.0",
+	"version": "0.3.1",
 	"description": "Vito's Pizzeria — a Pi distribution (Garfield-approved)",
 	"publishConfig": {
 		"access": "public",
@@ -37,15 +37,17 @@
 		"typescript": "^5.8.0"
 	},
 	"dependencies": {
-		"@vitos-pizza/agent-mode": "0.2.0",
-		"@vitos-pizza/keybindings": "0.2.0",
-		"@vitos-pizza/permission-system": "0.2.0",
-		"@vitos-pizza/question": "0.2.0",
-		"@vitos-pizza/session-title": "0.2.0",
-		"@vitos-pizza/subagents": "0.2.0",
-		"@vitos-pizza/todoist": "0.2.0",
-		"@vitos-pizza/ui-enhancements": "0.2.0",
-		"@vitos-pizza/websearch": "0.2.0"
+		"typebox": "^1.1.38",
+		"@vitos-pizza/agent-mode": "0.3.1",
+		"@vitos-pizza/keybindings": "0.3.1",
+		"@vitos-pizza/permission-system": "0.3.1",
+		"@vitos-pizza/question": "0.3.1",
+		"@vitos-pizza/session-title": "0.3.1",
+		"@vitos-pizza/settings-preset": "0.3.1",
+		"@vitos-pizza/subagents": "0.3.1",
+		"@vitos-pizza/todoist": "0.3.1",
+		"@vitos-pizza/ui-enhancements": "0.3.1",
+		"@vitos-pizza/websearch": "0.3.1"
 	},
 	"bundledDependencies": [
 		"@vitos-pizza/agent-mode",
@@ -61,14 +63,15 @@
 	"pi": {
 		"extensions": [
 			"node_modules/@vitos-pizza/permission-system/extensions",
+			"node_modules/@vitos-pizza/settings-preset/extensions",
 			"node_modules/@vitos-pizza/question/extensions",
 			"node_modules/@vitos-pizza/ui-enhancements/extensions",
-			"node_modules/@vitos-pizza/agent-mode/extensions",
 			"node_modules/@vitos-pizza/subagents/extensions",
 			"node_modules/@vitos-pizza/websearch/extensions",
 			"node_modules/@vitos-pizza/session-title/extensions",
-			"node_modules/@vitos-pizza/todoist/extensions",
-			"node_modules/@vitos-pizza/keybindings/extensions"
+			"node_modules/@vitos-pizza/keybindings/extensions",
+			"node_modules/@vitos-pizza/agent-mode/extensions",
+			"node_modules/@vitos-pizza/todoist/extensions"
 		],
 		"skills": [
 			"node_modules/@vitos-pizza/subagents/skills"
@@ -78,6 +81,7 @@
 		"type": "git",
 		"url": "git+https://github.com/xin2017338/vitos-pizza.git"
 	},
+	"bundleDependencies": [],
 	"files": [
 		"packages",
 		"assets",

package/packages/agent-mode/extensions/index.ts

@@ -19,10 +19,10 @@ import {
 } from "../src/mode-events.ts";
 import {
 	AGENT_MODES,
+	type AgentMode,
 	cycleAgentMode,
 	isAgentMode,
 	MODE_LABELS,
-	type AgentMode,
 } from "../src/modes.ts";
 import {
 	PLAN_INSTRUCTIONS,
@@ -62,16 +62,18 @@ export default function (pi: ExtensionAPI): void {
 		ctx.ui.notify(`Mode: ${mode}`, "info");
 	};
 
-	emitShortcutAction(pi.events, {
-		id: "agent-mode.cycle",
-		description: "Cycle agent mode (agent → plan → execute)",
-		handler: async (shortcutCtx) => {
-			const next = cycleAgentMode(currentMode);
-			await switchMode(shortcutCtx, next);
-		},
-	});
-
 	pi.on("session_start", async (_event, ctx) => {
+		// Register during session_start so @vitos-pizza/keybindings (loads later)
+		// already listens on vitos:shortcuts:register before we emit.
+		emitShortcutAction(pi.events, {
+			id: "agent-mode.cycle",
+			description: "Cycle agent mode (agent → plan → execute)",
+			handler: async (shortcutCtx) => {
+				const next = cycleAgentMode(currentMode);
+				await switchMode(shortcutCtx, next);
+			},
+		});
+
 		const mode = ensureAgentModePersisted(ctx.cwd, { emitReload });
 		publishMode(ctx, mode);
 	});

package/packages/agent-mode/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@vitos-pizza/agent-mode",
-	"version": "0.2.0",
+	"version": "0.3.1",
 	"description": "Centralized agent / plan / execute mode switching for Vito's Pizzeria",
 	"type": "module",
 	"license": "MIT",
@@ -13,15 +13,16 @@
 		"lint:fix": "biome check --write ."
 	},
 	"dependencies": {
-		"@vitos-pizza/keybindings": "0.2.0",
-		"@vitos-pizza/permission-system": "0.2.0"
+		"@vitos-pizza/keybindings": "0.3.1",
+		"@vitos-pizza/permission-system": "0.3.1"
 	},
 	"peerDependencies": {
 		"@earendil-works/pi-coding-agent": "*"
 	},
 	"pi": {
 		"requires": [
-			"@vitos-pizza/permission-system"
+			"@vitos-pizza/permission-system",
+			"@vitos-pizza/keybindings"
 		]
 	},
 	"devDependencies": {

package/packages/agent-mode/src/plan-instructions.ts

@@ -12,73 +12,37 @@ export const PLAN_MODE_TOOLS = [
 
 export const PLAN_INSTRUCTIONS = `## PLAN MODE
 
-You are in **plan** mode. Do not call write, edit, or bash on the main agent.
+This supersedes conflicting skill or prior guidance about implementing code or calling worker in this mode.
 
-**Planning workflow**
-- Mutually exclusive design choices (2–5 options) → always use the \`question\` tool; do not ask in free text
-- Produce an Implementation Plan and wait for user confirmation
-- Do **not** call worker or edit code directly in this mode
+**Hard rules**
+- Main agent: no write, edit, or bash. Do not call worker.
+- Produce an Implementation Plan and wait for confirmation.
+- To implement, the user switches via \`/mode execute\` or \`/mode agent\`.
 
-**When to explore**
-- Scout is **optional**. Skip it when the task is clear, scoped to known files, or the user already provided enough context
-- Call scout only when you need codebase recon (unfamiliar area, many files, unclear entry points)
-- Independent areas → parallel scouts via \`subagent({ tasks: [...] })\` to finish recon faster; then planner
-- For light lookups, the main agent may use read/grep/find/ls directly — no scout required
+**Clarification**
+- Prefer \`question\` for mutually exclusive choices (2–5 options); skip when context is enough. Available to main, scout, and planner.
+
+**Exploration**
+- Scout is optional — only for unfamiliar or multi-file recon. Parallel scouts via \`subagent({ tasks: [...] })\` for independent areas, then planner. Light lookups: use read/grep/find/ls yourself.
+- Web tools only for current external facts outside the repo.
 
 **Delegation**
-- Enough context already → \`subagent({ agent: "planner", task: "..." })\`
-- Need recon first → \`subagent({ chain: [{ agent: "scout", task: "..." }, { agent: "planner", task: "..." }] })\`
-- Multi-area recon → parallel scouts, then planner with the combined findings
+- Enough context → \`subagent({ agent: "planner", task })\`
+- Need recon → \`subagent({ chain: [scout, planner] })\`
 
-**After the plan (optional)**
-When helpful, add a short **Next** footer — Cursor-style suggested actions. Skip it if the next step is already obvious.
-- 2–3 bullets max; each one line
-- Concrete verbs the user can reply with (e.g. confirm → \`/mode execute\`, tweak scope)
-- No essays, no restating the whole plan
+**Optional Next footer** (2–3 one-line reply verbs) when the next step is not obvious:
 
-Example:
 \`\`\`
 **Next**
-- Confirm → \`/mode execute\` (or F6) to implement
-- Say what to change if the scope is off
-\`\`\`
-
-**To implement**, the user must switch to execute or agent mode (\`/mode execute\` or \`/mode agent\`).`;
+- Confirm → \`/mode execute\` (or Ctrl+. / Alt+M)
+- Say what to change if scope is off
+\`\`\``;
 
 export const PLAN_MODE_MESSAGE = `[PLAN MODE ACTIVE]
 
-You are in plan mode. The main agent must NOT write, edit, or run bash.
-
-1. Mutually exclusive choices → \`question\` tool (not free-text "which do you prefer?")
-2. Scout only if needed; parallel scouts (\`tasks: [...]\`) when exploring independent areas; otherwise go straight to planner
-3. Return the plan and wait; optionally a short **Next** footer (2–3 one-line actions) when helpful
-4. Do NOT call worker or make code changes
-
-Planner only (no scout):
-\`\`\`json
-{ "agent": "planner", "task": "Create an implementation plan for ..." }
-\`\`\`
-
-Scout → planner (when recon is needed):
-\`\`\`json
-{
-  "chain": [
-    { "agent": "scout", "task": "Map relevant code for this task" },
-    { "agent": "planner", "task": "Create an implementation plan from {previous}" }
-  ]
-}
-\`\`\`
-
-Parallel scouts (independent areas), then planner:
-\`\`\`json
-{
-  "tasks": [
-    { "agent": "scout", "task": "Map area A" },
-    { "agent": "scout", "task": "Map area B" }
-  ]
-}
-\`\`\`
-Then \`subagent({ agent: "planner", task: "Plan from the scout findings above" })\`.`;
+Read-only planning. No write/edit/bash/worker.
+Clarify with \`question\` when needed. Scout only if recon is required; otherwise planner.
+Return the plan and wait; optionally a short **Next** footer.`;
 
 export const PLAN_MODE_ENDED_MESSAGE = `[PLAN MODE ENDED]
 

package/packages/keybindings/extensions/index.ts

@@ -64,12 +64,7 @@ export function registerKeybindings(pi: ExtensionAPI): {
 } {
 	const registry = new ActionRegistry();
 	let state: KeybindingsState | null = null;
-
-	const unsubRegister = pi.events.on(SHORTCUTS_REGISTER, (payload) => {
-		if (!isShortcutAction(payload)) return;
-		registry.register(payload);
-	});
-	void unsubRegister;
+	let sessionCwd: string | undefined;
 
 	const rebind = (cwd: string) => {
 		const agentDir = getAgentDir();
@@ -81,10 +76,23 @@ export function registerKeybindings(pi: ExtensionAPI): {
 		return state;
 	};
 
+	const unsubRegister = pi.events.on(SHORTCUTS_REGISTER, (payload) => {
+		if (!isShortcutAction(payload)) return;
+		registry.register(payload);
+		// Rebind if session already started (covers either session_start order).
+		if (sessionCwd) rebind(sessionCwd);
+	});
+	void unsubRegister;
+
 	pi.on("session_start", async (_event, ctx) => {
+		sessionCwd = ctx.cwd;
 		rebind(ctx.cwd);
 	});
 
+	pi.on("session_shutdown", () => {
+		sessionCwd = undefined;
+	});
+
 	pi.registerCommand("vitos-shortcuts", {
 		description: "List vitos-pizza shortcut actions and key bindings",
 		handler: async (_args, ctx) => {

package/packages/keybindings/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@vitos-pizza/keybindings",
-	"version": "0.2.0",
+	"version": "0.3.1",
 	"description": "Centralized keyboard shortcut registry for Vito's Pizzeria Pi distribution",
 	"type": "module",
 	"license": "MIT",

package/packages/keybindings/presets/shortcuts.json

@@ -1,3 +1,3 @@
 {
-	"agent-mode.cycle": "f6"
+	"agent-mode.cycle": ["ctrl+.", "alt+m"]
 }

package/packages/permission-system/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@vitos-pizza/permission-system",
-	"version": "0.2.0",
+	"version": "0.3.1",
 	"description": "Permission enforcement for Vito's Pizzeria Pi distribution",
 	"type": "module",
 	"license": "MIT",
@@ -18,7 +18,7 @@
 		"lint:fix": "biome check --write ."
 	},
 	"dependencies": {
-		"@vitos-pizza/keybindings": "0.2.0",
+		"@vitos-pizza/keybindings": "0.3.1",
 		"tree-sitter-bash": "^0.25.1",
 		"web-tree-sitter": "^0.25.10",
 		"zod": "^3.25.76"

package/packages/question/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@vitos-pizza/question",
-	"version": "0.2.0",
+	"version": "0.3.1",
 	"description": "Structured ask-user question tool for Vito's Pizzeria Pi distribution",
 	"type": "module",
 	"license": "MIT",

package/packages/session-title/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@vitos-pizza/session-title",
-	"version": "0.2.0",
+	"version": "0.3.1",
 	"description": "Async session auto-naming for Vito's Pizzeria Pi distribution",
 	"type": "module",
 	"license": "MIT",
@@ -11,7 +11,7 @@
 		]
 	},
 	"dependencies": {
-		"@vitos-pizza/subagents": "0.2.0"
+		"@vitos-pizza/subagents": "0.3.1"
 	},
 	"scripts": {
 		"test": "vitest run",
@@ -30,7 +30,7 @@
 	"devDependencies": {
 		"@earendil-works/pi-ai": "*",
 		"@earendil-works/pi-coding-agent": "*",
-		"@vitos-pizza/subagents": "0.2.0",
+		"@vitos-pizza/subagents": "0.3.1",
 		"typebox": "*",
 		"vitest": "^3.2.4"
 	}

package/packages/settings-preset/extensions/index.ts

@@ -0,0 +1,53 @@
+/**
+ * @vitos-pizza/settings-preset — seed default Pi settings on first session start.
+ *
+ * Follows the same non-destructive pattern as the permission-system module:
+ * only writes to project-level `.pi/settings.json` when neither global nor
+ * project settings exist, so existing user configurations are never overwritten.
+ */
+
+import { copyFileSync, existsSync, mkdirSync } from "node:fs";
+import { dirname, join } from "node:path";
+import { fileURLToPath } from "node:url";
+import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
+import { getAgentDir } from "@earendil-works/pi-coding-agent";
+
+const moduleDir = dirname(fileURLToPath(import.meta.url));
+const presetsDir = join(moduleDir, "..", "presets");
+
+function getGlobalSettingsPath(agentDir: string): string {
+	return join(agentDir, "settings.json");
+}
+
+function getProjectSettingsPath(cwd: string): string {
+	return join(cwd, ".pi", "settings.json");
+}
+
+function ensureDefaultSettings(cwd: string, agentDir: string): void {
+	// Don't overwrite if the user already has project-level settings
+	const projectPath = getProjectSettingsPath(cwd);
+	if (existsSync(projectPath)) return;
+
+	// Don't overwrite if the user already has global settings
+	const globalPath = getGlobalSettingsPath(agentDir);
+	if (existsSync(globalPath)) return;
+
+	// Neither exists — seed the project with our preset
+	const presetPath = join(presetsDir, "settings.json");
+	if (!existsSync(presetPath)) return;
+
+	mkdirSync(dirname(projectPath), { recursive: true });
+	copyFileSync(presetPath, projectPath);
+}
+
+export default function (pi: ExtensionAPI) {
+	const agentDir = getAgentDir();
+	let seeded = false;
+
+	pi.on("session_start", async (_event, ctx) => {
+		if (seeded) return;
+		seeded = true;
+
+		ensureDefaultSettings(ctx.cwd, agentDir);
+	});
+}

package/packages/settings-preset/package.json

@@ -0,0 +1,19 @@
+{
+	"name": "@vitos-pizza/settings-preset",
+	"version": "0.3.1",
+	"description": "Default Pi settings for Vito's Pizzeria distribution — written on first session start",
+	"type": "module",
+	"license": "MIT",
+	"private": true,
+	"scripts": {
+		"typecheck": "tsc --noEmit",
+		"lint": "biome check .",
+		"lint:fix": "biome check --write ."
+	},
+	"peerDependencies": {
+		"@earendil-works/pi-coding-agent": "*"
+	},
+	"devDependencies": {
+		"@earendil-works/pi-coding-agent": "*"
+	}
+}

package/packages/settings-preset/presets/settings.json

@@ -0,0 +1,15 @@
+{
+	"quietStartup": true,
+	"hideThinkingBlock": true,
+	"session": {
+		"autoTitle": {
+			"enabled": true,
+			"fastRules": true,
+			"minCharsForLlm": 4,
+			"maxInputChars": 280,
+			"maxOutputTokens": 16,
+			"timeoutMs": 8000,
+			"maxTitleLength": 20
+		}
+	}
+}

package/packages/settings-preset/tsconfig.json

@@ -0,0 +1,11 @@
+{
+	"extends": "../../tsconfig.base.json",
+	"compilerOptions": {
+		"allowImportingTsExtensions": true,
+		"types": ["node"]
+	},
+	"include": [
+		"extensions/**/*.ts"
+	],
+	"exclude": ["node_modules"]
+}

package/packages/subagents/agents/planner.md

@@ -6,18 +6,13 @@ thinking: high
 systemPromptMode: replace
 ---
 
-You are a planning subagent.
+You are planner: turn requirements and code context into a concrete implementation plan. Do not change application code. You may `write` only to persist the plan document.
 
-Your job is to turn requirements and code context into a concrete implementation plan. Do not make code changes. Read, analyze, and write the plan only.
-
-Working rules:
-- Read the provided context before planning.
-- Read any additional code you need in order to make the plan concrete.
-- Name exact files whenever you can.
-- Prefer small, ordered, actionable tasks over vague phases.
-- Call out risks, dependencies, and anything that needs explicit validation.
-- If the task is underspecified, use the `question` tool to ask the user for clarification before guessing.
-- Mutually exclusive design choices (2–5 options) → always use `question`; never ask "which do you prefer?" in free text.
+Rules:
+- Read provided context (and more code as needed) before planning.
+- Name exact files; prefer small, ordered, actionable tasks; call out risks and dependencies.
+- Underspecified or mutually exclusive choices → use `question` before guessing.
+- Web tools only for current external facts outside the repo.
 
 Output format:
 
@@ -39,8 +34,9 @@ Numbered steps, each small and actionable.
 Which tasks depend on others.
 
 ## Risks
-Anything likely to go wrong, need clarification, or need careful verification.
+Anything likely to go wrong or need careful verification.
 
-Do **not** add a long "next steps" essay — the parent plan-mode agent may add a short **Next** footer when helpful.
+## 功能验收
+3–6 checkboxes of observable outcomes (yes/no verifiable).
 
-Keep the plan concrete. Another agent should be able to execute it without guessing what you meant.
+Keep the plan concrete enough for another agent to execute without guessing. The parent may add a short **Next** footer — do not write a long next-steps essay.

package/packages/subagents/agents/scout.md

@@ -1,27 +1,19 @@
 ---
 name: scout
 description: Fast codebase recon that returns compressed context for handoff
-tools: read, grep, find, ls, bash, write, question, web_search, web_read
+tools: read, grep, find, ls, question, web_search, web_read
 thinking: low
 systemPromptMode: replace
 ---
 
-You are a scouting subagent running inside pi.
+You are scout: fast codebase recon for handoff. Do not guess. Prefer targeted search over whole-file reads.
 
-Use the provided tools directly. Move fast, but do not guess. Prefer targeted search and selective reading over reading whole files unless the task clearly needs broader coverage.
+Return the minimum context another agent needs: entry points, key types/APIs, data flow, files likely to change, risks and open questions.
 
-Focus on the minimum context another agent needs in order to act:
-- relevant entry points
-- key types, interfaces, and functions
-- data flow and dependencies
-- files that are likely to need changes
-- constraints, risks, and open questions
-
-Working rules:
-- Use `grep`, `find`, `ls`, and `read` to map the area before diving deeper.
-- Use `bash` only for non-interactive inspection commands.
-- When you cite code, use exact file paths and line ranges.
-- When running solo, summarize what you found in the final response.
+Rules:
+- Cite exact file paths and line ranges.
+- Use `question` only when ambiguity blocks recon; prefer structured options over free-text choices.
+- Web tools only for current external facts outside the repo.
 
 Output format:
 
@@ -31,10 +23,10 @@ Output format:
 List exact files and line ranges.
 
 ## Key Code
-Include the critical types, interfaces, functions, and small code snippets that matter.
+Critical types, interfaces, functions, and small snippets that matter.
 
 ## Architecture
-Explain how the pieces connect.
+How the pieces connect.
 
 ## Start Here
-Name the first file another agent should open and why.
+First file another agent should open and why.

package/packages/subagents/agents/title.md

@@ -9,5 +9,6 @@ You analyze the user's intent together with the assistant's last reply and produ
 
 Rules:
 - Output only the title text — no quotes, no explanation, no markdown.
+- Match the language of the user's message.
 - If there is no clear task intent, output exactly `SKIP`.
 - Respect the maximum title length given in the task.