@ghyper9023/pi-dev-workflow 0.2.0 → 0.3.0

package/extensions/grill-me-agent.ts

@@ -0,0 +1,687 @@
+/**
+ * grill-me-agent.ts — 设计评审 (Grill) 和 PRD 生成的独立管理器
+ *
+ * 职责：
+ *   1. runGrillPhase()  — 启动 sub-agent 生成评审问题，TUI 逐题呈现（选项 + 自定义输入）
+ *   2. runPRDPhase()    — 启动 sub-agent 生成 PRD，保存到 pi-dev-output/pi-prd/
+ *
+ * 关键设计决策（修复 #2）：
+ *   sub-agent 通过 `write` 工具将评审问题写入临时文件，主进程事后读取。
+ *   不依赖从 NDJSON 响应文本中解析 JSON（多轮 tool-calling 场景不可靠）。
+ */
+
+import * as fs from "node:fs";
+import * as path from "node:path";
+import type { ExtensionAPI, ExtensionCommandContext } from "@earendil-works/pi-coding-agent";
+import { BorderedLoader, DynamicBorder } from "@earendil-works/pi-coding-agent";
+import { spawnSubagent, extractFinalOutput, type AgentDef } from "./sub-agents";
+export type { AgentDef };
+import {
+	Container,
+	SelectList,
+	Text,
+	Spacer,
+	type SelectItem,
+} from "@earendil-works/pi-tui";
+
+// ── Types ────────────────────────────────────────────────────
+
+/** A single grill question returned by the sub-agent. */
+export interface GrillQuestion {
+	id: number;
+	question: string;
+	options: string[];
+}
+
+/** Result of the grill phase. */
+export interface GrillResult {
+	/** Whether the user cancelled. */
+	cancelled: boolean;
+	/** The Q&A pairs collected. */
+	pairs: Array<{ question: string; answer: string }>;
+	/** Final enriched prompt (original prompt + all Q&A). */
+	enhancedPrompt: string;
+}
+
+/** Result of the PRD phase. */
+export interface PRDResult {
+	/** PRD content (Markdown). */
+	content: string;
+	/** Saved file path (relative to cwd). */
+	filePath: string;
+}
+
+// ── Output dirs ──────────────────────────────────────────────
+
+const DEV_OUTPUT_DIR = "pi-dev-output";
+const GRILL_DIRNAME = "pi-grill";
+const PRD_DIRNAME = "pi-prd";
+
+/** Ensure an output subdirectory exists and write .gitignore. */
+function ensureOutputDir(cwd: string, subdir: string): string {
+	const dir = path.join(cwd, DEV_OUTPUT_DIR, subdir);
+	fs.mkdirSync(dir, { recursive: true });
+	const gitignorePath = path.join(cwd, DEV_OUTPUT_DIR, ".gitignore");
+	try {
+		const existing = fs.readFileSync(gitignorePath, "utf-8").trim();
+		if (!existing.includes("*")) {
+			fs.writeFileSync(gitignorePath, "*\n!.gitignore\n");
+		}
+	} catch {
+		fs.writeFileSync(gitignorePath, "*\n!.gitignore\n");
+	}
+	return dir;
+}
+
+/** Generate a safe temp filename for grill output. */
+function grillOutputPath(cwd: string): string {
+	const dir = ensureOutputDir(cwd, GRILL_DIRNAME);
+	const ts = Date.now().toString(36);
+	return path.join(dir, `questions-${ts}.json`);
+}
+
+/** Generate a safe PRD filename. */
+function generatePrdFilename(moduleSuggestion: string): string {
+	const safe = moduleSuggestion
+		.replace(/[^a-zA-Z0-9\u4e00-\u9fff_-]/g, "-")
+		.replace(/-+/g, "-")
+		.replace(/^-|-$/g, "")
+		.slice(0, 40);
+	const ts = new Date().toISOString().slice(0, 10).replace(/-/g, "");
+	return `${safe || "feature"}-${ts}.md`;
+}
+
+// ── Sub-agent definitions ────────────────────────────────────
+
+/**
+ * Helper: build the system prompt suffix that tells the sub-agent to
+ * write results to a file via the `write` tool instead of putting JSON
+ * in its chat response.
+ */
+function writeToolPromptSuffix(outputFilePath: string): string {
+	return [
+		"",
+		"## Output via `write` tool (CRITICAL)",
+		"",
+		"Do NOT output the questions JSON in your chat response.",
+		"Instead, use the `write` tool to save the questions to a file.",
+		`Write to this exact path: ${outputFilePath}`,
+		"",
+		"The file content must be a valid JSON object with a single key \"questions\":",
+		'```json',
+		'{',
+		'  "questions": [',
+		'    {',
+		'      "id": 1,',
+		'      "question": "问题文本?",',
+		'      "options": ["选项A", "选项B", "选项C"]',
+		'    }',
+		'  ]',
+		'}',
+		'```',
+		"",
+		"After writing, you may include a brief summary in your chat response.",
+		"But the JSON MUST be in the file, NOT in the chat.",
+	].join("\n");
+}
+
+const GRILL_AGENT_DEF: AgentDef = {
+	name: "grill-agent",
+	description: "Design review agent — interviews the developer about a feature plan",
+	tools: ["read", "bash", "write"],
+	systemPrompt: [
+		"You are an expert design reviewer. Interview the developer relentlessly about every aspect of the feature plan until reaching shared understanding.",
+		"",
+		"## Rules",
+		"- For EACH question, provide recommended answer OPTIONS (a/b/c format) that the user can pick from",
+		"- Walk down each branch of the design tree, resolving dependencies between decisions one-by-one",
+		"- Be specific — refer to actual code modules, file paths, and architecture decisions",
+		"- Explore the codebase (use read/bash tools) when a question can be answered by looking at existing code",
+		"- Ask questions about: architecture, data flow, edge cases, security, testing, module boundaries, dependencies, error handling, performance, scalability",
+		"- If terminology conflicts with existing project glossary (CONTEXT.md), call it out",
+		"- Sharpen fuzzy language — propose precise canonical terms",
+		"- Stress-test scenarios with specific edge cases",
+		"- Cross-reference with existing code — surface contradictions",
+		"",
+		"## Quantity",
+		"Ask as many questions as needed to thoroughly review the design.",
+		"Do not artificially limit the number — cover architecture, data flow, edge cases, security, testing, module boundaries, dependencies, error handling, and more.",
+		"Typically 15-40 questions for a moderate feature.",
+		"",
+		"## Language",
+		"Questions and options should be in the same language as the feature request (default: Chinese).",
+	].join("\n"),
+	timeoutMs: 300_000, // 5 min
+};
+
+const PRD_AGENT_DEF: AgentDef = {
+	name: "prd-agent",
+	description: "PRD writer — synthesizes a PRD from conversation context",
+	tools: ["read", "bash"],
+	systemPrompt: [
+		"You are an expert product spec writer.",
+		"Your task is to create a PRD from the provided conversation context.",
+		"",
+		"## Rules",
+		"- Do NOT ask any questions — just synthesize what you already know",
+		"- Explore the repo to understand the current state of the codebase",
+		"- Use the project's domain vocabulary throughout",
+		"- Use the template below and output ONLY the Markdown content (no JSON wrapper, no preamble)",
+		"",
+		"## Template",
+		"",
+		"# {Feature Name} — PRD",
+		"",
+		"## Problem Statement",
+		"The problem that the user is facing, from the user's perspective.",
+		"",
+		"## Solution",
+		"The solution to the problem, from the user's perspective.",
+		"",
+		"## User Stories",
+		"A numbered list of user stories:",
+		"1. As an <actor>, I want a <feature>, so that <benefit>",
+		"",
+		"## Implementation Decisions",
+		"A list of implementation decisions including modules to build/modify, architectural decisions, schema changes, API contracts.",
+		"Do NOT include specific file paths or code snippets (may become outdated).",
+		"",
+		"## Testing Decisions",
+		"A description of what makes a good test, which modules will be tested, prior art.",
+		"",
+		"## Out of Scope",
+		"Things explicitly out of scope.",
+		"",
+		"## Further Notes",
+		"Any further notes about the feature.",
+	].join("\n"),
+	timeoutMs: 300_000,
+};
+
+// ── File-based question extraction ───────────────────────────
+
+/**
+ * Try to read and parse questions from the output file that the sub-agent
+ * was instructed to write via the `write` tool.
+ */
+function readQuestionsFromFile(filePath: string): GrillQuestion[] {
+	try {
+		if (!fs.existsSync(filePath)) return [];
+		const raw = fs.readFileSync(filePath, "utf-8").trim();
+		if (!raw) return [];
+		const parsed = JSON.parse(raw);
+		// Support both { questions: [...] } and raw [...]
+		const items = Array.isArray(parsed) ? parsed : parsed.questions;
+		if (!Array.isArray(items)) return [];
+		return items
+			.filter((q: any) => q && typeof q.question === "string" && Array.isArray(q.options) && q.options.length > 0)
+			.map((q: any, i: number) => ({
+				id: q.id ?? i + 1,
+				question: q.question,
+				options: q.options,
+			}));
+	} catch {
+		return [];
+	}
+}
+
+/**
+ * Parse sub-agent output into a list of GrillQuestions.
+ * Used as fallback when the file-based approach didn't produce results.
+ */
+export function parseGrillQuestions(raw: string): GrillQuestion[] {
+	if (!raw || !raw.trim()) return [];
+
+	// Strategy 1: find any {} JSON, deep-search for "questions" key
+	const jsonMatch = raw.match(/\{[\s\S]*\}/);
+	if (jsonMatch) {
+		const candidate = jsonMatch[0]
+			.replace(/^\s*```(?:json)?\s*/, "")
+			.replace(/\s*```\s*$/, "");
+		try {
+			const parsed = JSON.parse(candidate);
+			const found = deepFindQuestions(parsed);
+			if (found && found.length > 0) {
+				return found
+					.filter((q) => q.question && q.options && q.options.length > 0)
+					.map((q, i) => ({
+						id: q.id ?? i + 1,
+						question: q.question!,
+						options: q.options!,
+					}));
+			}
+		} catch {
+			// fall through
+		}
+	}
+
+	// Strategy 2: try to find a question-like JSON array directly
+	const direct = extractQuestionArray(raw);
+	if (direct.length > 0) {
+		return direct.map((q, i) => ({ id: i + 1, question: q.question, options: q.options }));
+	}
+
+	// Strategy 3: extract from markdown code block
+	const codeBlock = raw.match(/```(?:json)?\s*\n?([\s\S]*?)```/);
+	if (codeBlock) {
+		return parseGrillQuestions(codeBlock[1]);
+	}
+
+	return [];
+}
+
+function deepFindQuestions(obj: unknown): Array<{ id?: number; question?: string; options?: string[] }> | null {
+	if (!obj || typeof obj !== "object") return null;
+	if (Array.isArray(obj)) {
+		for (const item of obj) {
+			const found = deepFindQuestions(item);
+			if (found) return found;
+		}
+		return null;
+	}
+	const record = obj as Record<string, unknown>;
+	if ("questions" in record && Array.isArray(record.questions)) {
+		return record.questions as Array<{ id?: number; question?: string; options?: string[] }>;
+	}
+	for (const val of Object.values(record)) {
+		const found = deepFindQuestions(val);
+		if (found) return found;
+	}
+	return null;
+}
+
+function extractQuestionArray(raw: string): Array<{ question: string; options: string[] }> {
+	const arrayMatch = raw.match(/\[\s*\{[\s\S]*?"question"[\s\S]*?"options"[\s\S]*?\}\s*\]/);
+	if (!arrayMatch) return [];
+	try {
+		const items = JSON.parse(arrayMatch[0]);
+		if (!Array.isArray(items)) return [];
+		return items.filter((item: unknown): item is { question: string; options: string[] } => {
+			if (!item || typeof item !== "object") return false;
+			const r = item as Record<string, unknown>;
+			return typeof r.question === "string" && Array.isArray(r.options) && r.options.length > 0;
+		});
+	} catch {
+		return [];
+	}
+}
+
+// ── Grill Phase ──────────────────────────────────────────────
+
+/**
+ * Options for customizing the grill phase.
+ */
+export interface GrillOptions {
+	agentDef?: AgentDef;
+	title?: string;
+	description?: string;
+	questionTitle?: string;
+	loaderLabel?: string;
+}
+
+/**
+ * Run the grill phase:
+ * 1. Confirm with user
+ * 2. Call sub-agent → sub-agent writes questions to a file via `write` tool
+ * 3. Read questions from the file
+ * 4. Show each question as TUI SelectList + custom input option
+ * 5. Collect all Q&A pairs
+ * 6. Return enhanced prompt
+ */
+export async function runGrillPhase(
+	assembledPrompt: string,
+	ctx: ExtensionCommandContext,
+	options?: GrillOptions,
+): Promise<GrillResult> {
+	const defaultResult: GrillResult = {
+		cancelled: false,
+		pairs: [],
+		enhancedPrompt: assembledPrompt,
+	};
+
+	const agentDef = options?.agentDef ?? GRILL_AGENT_DEF;
+	const confirmTitle = options?.title ?? "🔍 设计方案评审";
+	const confirmDesc = options?.description ?? "是否进入设计评审 (Grill) 模式？\nAI 会从架构、数据流、边界条件、安全等多个维度挑战你的设计。";
+	const qTitlePrefix = options?.questionTitle ?? "设计方案评审";
+	const loaderLabel = options?.loaderLabel ?? "🧠 AI 正在分析代码并生成评审问题...";
+
+	// ── Step 1: Confirm entering grill mode ──────────────────
+	const enterGrill = await ctx.ui.confirm(confirmTitle, confirmDesc);
+	if (!enterGrill) {
+		return { ...defaultResult, cancelled: true };
+	}
+
+	// ── Step 2: Prepare output file + enhanced prompt ─────────
+	const outputFilePath = grillOutputPath(ctx.cwd);
+	const enhancedPrompt = [
+		assembledPrompt,
+		writeToolPromptSuffix(outputFilePath),
+	].join("\n\n");
+
+	// ── Step 3: Call sub-agent with BorderedLoader ───────────
+	const questions = await ctx.ui.custom<GrillQuestion[]>((tui, theme, _kb, done) => {
+		const loader = new BorderedLoader(tui, theme, loaderLabel);
+		loader.onAbort = () => done([]);
+
+		spawnSubagent(
+			agentDef,
+			enhancedPrompt,
+			ctx.cwd,
+			loader.signal,
+			undefined,
+			(progress) => { loader.setText(`🧠 ${progress.slice(0, 60)}`); },
+		)
+			.then((result) => {
+				// Primary: read from file (sub-agent wrote via `write` tool)
+				let qs = readQuestionsFromFile(outputFilePath);
+				// Fallback: parse from NDJSON response text
+				if (qs.length === 0) {
+					const output = extractFinalOutput(result.output);
+					qs = parseGrillQuestions(output);
+				}
+				done(qs);
+			})
+			.catch(() => {
+				// On error, still try to read file (sub-agent may have written before error)
+				const qs = readQuestionsFromFile(outputFilePath);
+				done(qs);
+			});
+
+		return loader;
+	});
+
+	// ── Step 4: Clean up temp file ───────────────────────────
+	try { fs.unlinkSync(outputFilePath); } catch { /* ignore */ }
+
+	// ── Step 4b: Retry dialog if no questions generated ──────
+	if (questions.length === 0) {
+		const choice = await ctx.ui.select(
+			"⚠️ AI 未能成功生成评审问题",
+			[
+				"🔄 重新尝试生成评审问题",
+				"⏭️ 跳过 Grill，直接发送 Prompt",
+				"❌ 取消 (Esc)",
+			],
+		);
+
+		switch (choice) {
+			case "🔄 重新尝试生成评审问题": {
+				const retryPath = grillOutputPath(ctx.cwd);
+				const retryPrompt = [assembledPrompt, writeToolPromptSuffix(retryPath)].join("\n\n");
+				questions = await ctx.ui.custom<GrillQuestion[]>((tui, theme, _kb, done) => {
+					const loader = new BorderedLoader(tui, theme, loaderLabel);
+					loader.onAbort = () => done([]);
+					spawnSubagent(agentDef, retryPrompt, ctx.cwd, loader.signal, undefined)
+						.then((r) => {
+							let qs = readQuestionsFromFile(retryPath);
+							if (qs.length === 0) qs = parseGrillQuestions(extractFinalOutput(r.output));
+							done(qs);
+						})
+						.catch(() => done([]));
+					return loader;
+				});
+				try { fs.unlinkSync(retryPath); } catch { /* ignore */ }
+				if (questions.length === 0) {
+					ctx.ui.notify("⚠️ 再次尝试仍然失败，跳过 Grill 阶段", "warning");
+					return defaultResult;
+				}
+				break;
+			}
+			case "⏭️ 跳过 Grill，直接发送 Prompt":
+				ctx.ui.notify("⏭️ 已跳过 Grill 阶段", "info");
+				return defaultResult;
+			case "❌ 取消 (Esc)":
+			default:
+				ctx.ui.notify("❌ 操作已取消", "warning");
+				return { ...defaultResult, cancelled: true };
+		}
+	}
+
+	ctx.ui.notify(`✅ AI 生成了 ${questions.length} 个评审问题`, "success");
+
+	// ── Step 5: TUI — present questions one by one ───────────
+	const pairs: Array<{ question: string; answer: string }> = [];
+
+	for (let idx = 0; idx < questions.length; idx++) {
+		const q = questions[idx];
+		const answer = await showQuestionTUI(ctx, q, idx + 1, questions.length, qTitlePrefix);
+
+		if (answer === null) {
+			ctx.ui.notify("❌ 评审已取消", "warning");
+			return { ...defaultResult, cancelled: true, pairs };
+		}
+
+		pairs.push({ question: q.question, answer });
+	}
+
+	// ── Step 6: Assemble enhanced prompt ─────────────────────
+	const qaBlock = pairs
+		.map((p, i) => `[评审问题 ${i + 1}]\n问题: ${p.question}\n回答: ${p.answer}`)
+		.join("\n\n");
+
+	const finalEnhancedPrompt = [
+		assembledPrompt,
+		"",
+		"---",
+		"## 设计评审记录",
+		"",
+		"以下是在开发前进行的设计评审问答，所有决策已确认：",
+		"",
+		qaBlock,
+	].join("\n");
+
+	ctx.ui.notify(`✅ 评审完成，共 ${pairs.length} 道问题`, "success");
+
+	return {
+		cancelled: false,
+		pairs,
+		enhancedPrompt: finalEnhancedPrompt,
+	};
+}
+
+/** Show a single grill question as TUI SelectList + custom input option. */
+async function showQuestionTUI(
+	ctx: ExtensionCommandContext,
+	q: GrillQuestion,
+	currentIndex: number,
+	totalCount: number,
+	titlePrefix = "设计方案评审",
+): Promise<string | null> {
+	const selectItems: SelectItem[] = q.options.map((opt, i) => ({
+		value: `opt-${i}`,
+		label: `(${String.fromCharCode(97 + i)}) ${opt}`,
+	}));
+	selectItems.push({
+		value: "__custom__",
+		label: "✏️  自定义输入",
+		description: "输入你自己的回答，不受选项限制",
+	});
+
+	const title = `${titlePrefix} (问题 ${currentIndex}/${totalCount})`;
+
+	const value = await ctx.ui.custom<string | null>((tui, theme, _kb, done) => {
+		const container = new Container();
+		container.addChild(new DynamicBorder((s: string) => theme.fg("accent", s)));
+		container.addChild(new Text(theme.fg("accent", theme.bold(`  ${title}`)), 0, 0));
+		container.addChild(new Spacer(1));
+		container.addChild(new Text(theme.fg("text", `  ${q.question}`), 0, 0));
+		container.addChild(new Spacer(1));
+
+		const visibleCount = Math.min(selectItems.length + 1, 12);
+		const selectList = new SelectList(selectItems, visibleCount, {
+			selectedPrefix: (s) => theme.fg("accent", s),
+			selectedText: (s) => theme.fg("accent", s),
+			description: (s) => theme.fg("muted", s),
+			scrollInfo: (s) => theme.fg("dim", s),
+			noMatch: (s) => theme.fg("warning", s),
+		});
+		selectList.onSelect = (item) => done(item.value);
+		selectList.onCancel = () => done(null);
+		container.addChild(selectList);
+
+		container.addChild(new Spacer(1));
+		container.addChild(
+			new Text(theme.fg("dim", "  ↑↓ 导航 • Enter 选择 • Esc 取消全部评审"), 0, 0),
+		);
+		container.addChild(new DynamicBorder((s: string) => theme.fg("accent", s)));
+
+		return {
+			render: (w) => container.render(w),
+			invalidate: () => container.invalidate(),
+			handleInput: (data) => {
+				selectList.handleInput(data);
+				tui.requestRender();
+			},
+		};
+	});
+
+	if (value === null) return null;
+	if (value === "__custom__") {
+		const custom = await ctx.ui.input("✏️ 自定义回答", {
+			placeholder: "输入你的回答内容（Esc 取消本题，回到选项）",
+			required: false,
+		});
+		if (custom === undefined) return showQuestionTUI(ctx, q, currentIndex, totalCount);
+		return custom.trim() || "(空)";
+	}
+
+	const optIndex = parseInt(value.replace("opt-", ""), 10);
+	return q.options[optIndex] || value;
+}
+
+// ── PRD Phase ────────────────────────────────────────────────
+
+/**
+ * Run the PRD phase:
+ * 1. Ask user if they want to create a PRD
+ * 2. Call sub-agent → gets PRD Markdown
+ * 3. Save to pi-dev-output/pi-prd/<name>.md
+ * 4. Ask if user wants to start development
+ */
+export async function runPRDPhase(
+	context: string,
+	moduleHint: string,
+	pi: ExtensionAPI,
+	ctx: ExtensionCommandContext,
+): Promise<PRDResult | null> {
+	const wantPrd = await ctx.ui.confirm(
+		"📋 创建 PRD",
+		"是否为此功能创建 PRD 文档？\nPRD 将保存到 pi-dev-output/pi-prd/ 目录。",
+	);
+	if (!wantPrd) return null;
+
+	const prdContent = await ctx.ui.custom<string | null>((tui, theme, _kb, done) => {
+		const loader = new BorderedLoader(tui, theme, "📝 AI 正在生成 PRD 文档...");
+		loader.onAbort = () => done(null);
+
+		const prdTask = [
+			"请根据以下上下文生成一份 PRD（产品需求文档）。",
+			"输出格式必须是完整的 Markdown 文档。",
+			"",
+			"=== 上下文 ===",
+			context,
+		].join("\n");
+
+		spawnSubagent(PRD_AGENT_DEF, prdTask, ctx.cwd, loader.signal)
+			.then((result) => {
+				const output = extractFinalOutput(result.output);
+				done(output && output.length >= 50 ? output : null);
+			})
+			.catch(() => done(null));
+
+		return loader;
+	});
+
+	if (!prdContent) {
+		ctx.ui.notify("⚠️ PRD 生成失败", "error");
+		return null;
+	}
+
+	const prdDir = ensureOutputDir(ctx.cwd, PRD_DIRNAME);
+	const filename = generatePrdFilename(moduleHint);
+	const filePath = path.join(DEV_OUTPUT_DIR, PRD_DIRNAME, filename);
+	const fullPath = path.join(prdDir, filename);
+	fs.writeFileSync(fullPath, prdContent, "utf-8");
+	ctx.ui.notify(`✅ PRD 已保存到 ${filePath}`, "success");
+
+	await askDevelopmentStart(pi, ctx, prdContent, filePath);
+	return { content: prdContent, filePath };
+}
+
+async function askDevelopmentStart(
+	pi: ExtensionAPI,
+	ctx: ExtensionCommandContext,
+	prdContent: string,
+	prdFilePath: string,
+): Promise<void> {
+	const choice = await ctx.ui.select(
+		"🚀 是否开始开发？",
+		[
+			"是 — 根据 PRD 开始开发",
+			"否 — 稍后手动开始",
+			"✏️ 自定义开发指令",
+		],
+	);
+	if (!choice) return;
+
+	switch (choice) {
+		case "是 — 根据 PRD 开始开发": {
+			const devMsg = [
+				"请根据以下 PRD 文档开始开发：",
+				"",
+				`PRD 文件: \`${prdFilePath}\``,
+				"",
+				"--- PRD 全文 ---",
+				prdContent,
+				"",
+				"---",
+				"",
+				"请按照上述 PRD 逐步实现。先分析代码库结构，给出实施计划，确认后再编写代码。",
+			].join("\n");
+			pi.sendUserMessage(devMsg);
+			ctx.ui.notify("🚀 已发送开发指令给主代理", "success");
+			break;
+		}
+		case "否 — 稍后手动开始":
+			ctx.ui.notify(`📋 PRD 已保存在 ${prdFilePath}，可随时手动引用`, "info");
+			break;
+		default: {
+			const customMsg = await ctx.ui.input("✏️ 自定义开发指令", {
+				placeholder: "输入你的开发指令（将结合 PRD 一起发送给主代理）",
+				required: false,
+			});
+			if (customMsg === undefined) return askDevelopmentStart(pi, ctx, prdContent, prdFilePath);
+			const finalMsg = customMsg.trim()
+				? [
+					`自定义开发指令: ${customMsg.trim()}`,
+					"",
+					`PRD 文件: \`${prdFilePath}\``,
+					"",
+					"--- PRD 全文 ---",
+					prdContent,
+				].join("\n")
+				: [
+					"请根据以下 PRD 文档开始开发：",
+					"",
+					`PRD 文件: \`${prdFilePath}\``,
+					"",
+					"--- PRD 全文 ---",
+					prdContent,
+				].join("\n");
+			pi.sendUserMessage(finalMsg);
+			ctx.ui.notify("🚀 已发送自定义开发指令给主代理", "success");
+			break;
+		}
+	}
+}
+
+// ── Extension factory (required by pi extension loader) ─────
+//
+// This file lives in extensions/ so pi will attempt to load it.
+// The default export satisfies the loader; the real functionality
+// is consumed by dev-prompts.ts via named imports.
+//
+export default function (_pi: ExtensionAPI) {
+	// grill-me-agent is a helper module, not a standalone extension.
+	// It is imported by dev-prompts.ts to provide grill + PRD phases.
+}