hyacinth-ai 0.9.6

package/docs/diagnostics/2026-06-08-config-prompt-guidance.md

@@ -0,0 +1,67 @@
+# 配置提示词增强记录
+
+日期: 2026-06-08
+
+## 背景
+
+之前在 release 环境中尝试让 Agent 配置飞书渠道时，模型没有明确知道应该修改哪个配置文件，也容易把渠道配置当成普通运行时配置处理。
+
+排查后确认：
+
+- `src/prompts/environment/feishu.md` 已经包含飞书配置细节，但它不是默认注入上下文。
+- `src/prompts/tools/framework-capabilities.md` 是默认注入的框架能力提示词，适合放通用配置指导。
+- `src/prompts/environment/environment.md` 是默认环境段，适合放配置文件路径索引。
+
+## 修改内容
+
+### framework-capabilities.md
+
+文件: `src/prompts/tools/framework-capabilities.md`
+
+新增了两类配置指导：
+
+1. 运行时配置
+   - 明确 `get_config` / `update_config` / `reset_config` / `config_schema` 适合修改 RuntimeConfigCenter 已支持的路径。
+   - 补充常见路径，例如 `provider.active`、`provider.<provider>.model`、`session.maxTurns`、`session.maxContext`、`safety.requireConfirmation`、`tools.disabled`、`skills.disabled`、`models.assessment`、`logging.level`、`hotReload.*` 等。
+
+2. 项目文件配置
+   - 明确部分配置不应优先使用 `update_config`，而应读取并写回项目 `.agent/` 下的 JSON 文件。
+   - 覆盖渠道、MCP、工具包、Provider 元数据、本地模型、本地 Provider、子 Agent、身份提示词等配置入口。
+   - 为渠道配置增加专门说明，明确飞书/Lark 等渠道写入项目级 `.agent/config.json` 的 `channels` 字段。
+   - 强调修改 JSON 时必须保留已有字段，只合并用户要求的部分。
+
+### environment.md
+
+文件: `src/prompts/environment/environment.md`
+
+新增默认环境路径索引：
+
+- 全局身份提示词: `~/.agent/prompts/persona/*.md`
+- 渠道配置: 项目级 `{{cwd}}/.agent/config.json` 的 `channels` 字段
+- MCP 配置: `{{cwd}}/.agent/mcp.json`
+- 工具包配置: `{{cwd}}/.agent/tool-bundles.json`
+- Provider 元数据: `{{cwd}}/.agent/providers.json`
+- 本地模型配置: `{{cwd}}/.agent/models.json` 或 `{{cwd}}/.agent/local-models.json`
+
+## 预期效果
+
+模型在用户要求修改框架设置时，应先判断目标属于哪类配置：
+
+- 如果路径存在于 `config_schema`，优先使用 `update_config`。
+- 如果是渠道、MCP、工具包、Provider 元数据、本地模型等项目文件配置，则使用 `read` + `write` 修改对应 JSON。
+- 配置飞书/Lark 渠道时，优先修改当前工作区 `.agent/config.json` 的 `channels.feishu`，不要误用 `update_config`。
+- 修改前应告知用户当前值和计划改成的值，且不应把 API Key 写入不合适的位置。
+
+## 验证
+
+已执行：
+
+```powershell
+corepack pnpm run build
+```
+
+结果：
+
+- TypeScript 编译通过。
+- `src/prompts` 已同步复制到 `dist/prompts`。
+- 已确认 `dist/prompts/tools/framework-capabilities.md` 中包含新增的运行时配置、项目文件配置和渠道配置说明。

package/docs/superpowers/plans/2026-06-08-global-bootstrap-mode.md

@@ -0,0 +1,89 @@
+# Global Bootstrap Mode Implementation Plan
+
+> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
+
+**Goal:** Replace session/file-deletion bootstrap with a global first-run `bootstrap` mode that writes persona files once and then stops injecting.
+
+**Architecture:** Reuse the existing `ModeManager` lifecycle and Zone 5 `mode-injection` path. Add a global persona bootstrap store, a `bootstrap` mode, and a dedicated `bootstrap_mark` tool; startup code activates the mode only when global persona state is incomplete.
+
+**Tech Stack:** TypeScript, Node.js ESM, Vitest, existing `ModeDefinition`, `ConfigManager`, and prompt loader patterns.
+
+---
+
+### Task 1: Global Persona Store
+
+**Files:**
+- Modify: `src/setup/persona-bootstrap.ts`
+- Test: `src/setup/persona-bootstrap.test.ts`
+
+- [ ] Add tests for global directory resolution, template creation, validation, and completion state.
+- [ ] Refactor `DEFAULT_PERSONA_DIR` to resolve under `~/.agent/prompts/persona` by default.
+- [ ] Add `ensureGlobalPersonaFiles()`, `validatePersonaFiles()`, `isBootstrapComplete()`, and `markBootstrapComplete()`.
+- [ ] Keep compatibility wrappers for existing callers where practical.
+- [ ] Run `pnpm test src/setup/persona-bootstrap.test.ts`.
+
+### Task 2: Bootstrap Mode
+
+**Files:**
+- Create: `src/modes/bootstrap.mode.ts`
+- Modify: `src/modes/index.ts`
+- Create: `src/prompts/modes/bootstrap.md`
+- Test: `src/modes/bootstrap.mode.test.ts`
+
+- [ ] Add a failing test that `bootstrap` renders mode guidance and stays active before completion.
+- [ ] Add a failing test that `bootstrap_mark({ action: "complete" })` only completes after validation passes.
+- [ ] Implement `createBootstrapMode(personaDir)` using the existing `ModeDefinition` interface.
+- [ ] Add prompt instructions for natural first-run identity collection and persona file writing.
+- [ ] Export the mode from `src/modes/index.ts`.
+- [ ] Run `pnpm test src/modes/bootstrap.mode.test.ts`.
+
+### Task 3: Bootstrap Mark Tool
+
+**Files:**
+- Create: `src/tools/bootstrap.ts`
+- Modify: `src/gateway/factory.ts`
+- Test: `src/tools/bootstrap.test.ts`
+
+- [ ] Add a failing test for `bootstrap_mark` returning progress and completion errors.
+- [ ] Implement a small tool that delegates `progress` and `complete` actions to `ModeManager.dispatchToolCall()`.
+- [ ] Register `bootstrap_mark` in `createAgent()` after `ModeManager` exists.
+- [ ] Ensure `bootstrap_mark` is available during bootstrap even when tool bundles are active.
+- [ ] Run `pnpm test src/tools/bootstrap.test.ts`.
+
+### Task 4: Startup Integration
+
+**Files:**
+- Modify: `src/gateway/cli.ts`
+- Modify: `src/gateway/tui.ts`
+- Modify: `src/gateway/factory.ts`
+- Modify: `src/orchestrator/loop.ts`
+- Test: focused integration/unit tests if existing gateway tests permit
+
+- [ ] Ensure both TUI and non-TUI paths call the same global persona status helpers.
+- [ ] Register `bootstrap` mode in `createAgent()` and auto-activate it when global state is incomplete.
+- [ ] Trigger an automatic first model turn when bootstrap is active.
+- [ ] Remove reliance on deleting `BOOTSTRAP.md` as the completion signal.
+- [ ] Keep `startBootstrap()` as a compatibility wrapper around the new active mode if needed.
+- [ ] Run `pnpm test` or the narrowest available test set.
+
+### Task 5: Prompt Loading and Backward Compatibility
+
+**Files:**
+- Modify: `src/context/section-resolver.ts`
+- Modify: `src/prompts/persona/BOOTSTRAP.md`
+- Test: covered by previous tests or add resolver test if available
+
+- [ ] Prefer project persona override, then global persona files, then built-in templates for `SOUL/IDENTITY/USER`.
+- [ ] Stop using `persona/BOOTSTRAP.md` as a stateful bootstrap prompt.
+- [ ] Turn `BOOTSTRAP.md` into a legacy note or leave it unused for compatibility.
+- [ ] Verify compose output includes bootstrap only through `mode-injection`.
+
+### Task 6: Final Verification
+
+**Files:**
+- All changed files
+
+- [ ] Run `pnpm test`.
+- [ ] Run `pnpm build`.
+- [ ] Inspect changed files for accidental secrets or home-directory writes beyond intended global persona defaults.
+- [ ] Summarize final behavior and any test gaps.

package/docs/superpowers/specs/2026-06-08-global-bootstrap-mode-design.md

@@ -0,0 +1,174 @@
+# Global Bootstrap Mode Design
+
+## Why
+
+The first-run identity bootstrap is a staged task, but the current implementation treats it like a conditional prompt plus a file-deletion signal. This makes it easy to retrigger per session, skip in some startup paths, or complete without verifying that persona files were actually written.
+
+## What Changes
+
+- Add a dedicated `bootstrap` mode alongside `plan`, `spec`, and `todo`.
+- Store bootstrap state globally, not in a session directory.
+- Move bootstrap instructions from `persona/BOOTSTRAP.md` semantics into a mode prompt such as `prompts/modes/bootstrap.md`.
+- Trigger bootstrap automatically only when the global persona state is incomplete.
+- Complete bootstrap only after required global persona files are written and validated.
+- Stop injecting bootstrap guidance immediately after completion.
+
+## Scope
+
+In scope:
+
+- Global persona bootstrap lifecycle.
+- Startup detection and automatic activation.
+- Mode prompt injection through the existing `mode-injection` Zone 5 path.
+- Completion state persisted in global persona state.
+- TUI and non-TUI startup consistency.
+
+Out of scope:
+
+- Provider installation and API key setup.
+- Changing normal `plan`, `spec`, or `todo` behavior.
+- Project-specific persona profiles beyond optional future override support.
+
+## Storage
+
+Use a global persona directory as the default identity source:
+
+```text
+~/.agent/
+  prompts/
+    persona/
+      SOUL.md
+      IDENTITY.md
+      USER.md
+      .state/
+        persona-state.json
+```
+
+The project-level `.agent/prompts/persona` directory may remain as an override mechanism, but first-run identity bootstrap writes to the global persona directory.
+
+The state file records at least:
+
+```json
+{
+  "version": 1,
+  "bootstrapSeededAt": "ISO timestamp",
+  "setupCompletedAt": "ISO timestamp"
+}
+```
+
+`setupCompletedAt` is the source of truth for completion. Deleting a prompt file is not a completion signal.
+
+## Startup Flow
+
+1. Load normal install/config state.
+2. Resolve the global persona directory.
+3. Ensure global `SOUL.md`, `IDENTITY.md`, and `USER.md` exist.
+4. Read global persona state.
+5. If `setupCompletedAt` exists, start normally.
+6. If `setupCompletedAt` is missing, activate `bootstrap` mode before the first model turn.
+7. The first model turn is initiated automatically so the Agent greets the user and starts collecting identity preferences.
+8. Bootstrap mode remains active across turns until completion.
+9. On completion, write state, deactivate the mode, and continue normal conversation without bootstrap injection.
+
+## Bootstrap Mode Behavior
+
+`bootstrap` is a framework mode, not a normal task mode users are expected to activate for daily work.
+
+It should:
+
+- Prompt the Agent to greet the user naturally.
+- Collect Agent identity fields: name, role, style, boundaries, optional symbol.
+- Collect user fields: name, preferred address, timezone, background, preferences.
+- Collect collaboration preferences: initiative level, confirmation boundaries, coding style.
+- Ask follow-up questions only when required information is missing.
+- Write persona files with existing tools or a dedicated persona write helper.
+- Mark completion only after validation passes.
+
+## Completion Contract
+
+Bootstrap can complete when all of the following are true:
+
+- `IDENTITY.md` no longer contains the untouched template placeholders.
+- `USER.md` no longer contains the untouched template placeholders.
+- `SOUL.md` exists and remains usable as the long-term style/behavior guide.
+- The user has either explicitly confirmed the captured identity information or the Agent has summarized it and received no correction in the final bootstrap turn.
+
+After completion:
+
+- Write `setupCompletedAt`.
+- Deactivate `bootstrap` mode.
+- Do not inject `bootstrap` prompt again in new sessions.
+
+## Components
+
+### `BootstrapMode`
+
+New mode definition implementing the existing `ModeDefinition` interface:
+
+- `createState()` initializes global persona paths and required fields.
+- `renderForInjection()` renders `prompts/modes/bootstrap.md`.
+- `handleToolCall()` accepts progress markers or a completion marker.
+- `isComplete()` returns true only after validation.
+
+### `PersonaBootstrapStore`
+
+Small global store for:
+
+- Resolving the global persona directory.
+- Ensuring template files exist.
+- Reading and writing persona state.
+- Validating whether persona files are still templates.
+- Marking bootstrap complete.
+
+Existing `persona-bootstrap.ts` can be refactored into this role.
+
+### Startup Integration
+
+Both TUI and non-TUI creation paths should pass the same global persona state into `createAgent`.
+
+If bootstrap is incomplete, `createAgent` or the gateway activates `bootstrap` mode before the first turn. The activation should not depend on a session-specific file.
+
+### Prompt Loading
+
+Long-term persona loading should prefer:
+
+1. Project override, if explicitly present and intended.
+2. Global persona files.
+3. Built-in templates as fallback.
+
+Bootstrap mode prompt should be loaded from `prompts/modes/bootstrap.md`, not from `persona/BOOTSTRAP.md`.
+
+## Error Handling
+
+- If persona files cannot be written, keep bootstrap active and report the write failure.
+- If validation fails, keep bootstrap active and inject the missing fields in the next turn.
+- If the user exits mid-bootstrap, keep state incomplete so the next session resumes bootstrap.
+- If global state says complete but persona files are missing, recreate missing files and mark bootstrap incomplete.
+
+## Testing
+
+Unit tests:
+
+- Global store creates missing persona files.
+- Incomplete state activates bootstrap.
+- Complete state does not activate bootstrap.
+- Template files fail validation.
+- Filled files pass validation.
+
+Integration tests:
+
+- First TUI startup activates bootstrap once.
+- First non-TUI startup follows the same activation rule.
+- Completing bootstrap writes global state and stops injection.
+- New session after completion does not reinject bootstrap.
+
+## Completion Tool
+
+Use a dedicated `bootstrap_mark` tool instead of reusing `task_mark`.
+
+The tool should support at least:
+
+- `bootstrap_mark({ action: "progress", message })` for recording what has been collected.
+- `bootstrap_mark({ action: "complete" })` for requesting completion after files have been written.
+
+The mode still performs validation before accepting completion. If validation fails, the tool result explains what is missing and bootstrap mode remains active.

package/docs/user-guide.md

@@ -0,0 +1,226 @@
+# User Guide
+
+## Installation
+
+```bash
+# Install dependencies
+pnpm install
+
+# Build
+pnpm build
+
+# Link CLI globally
+npm link  # or: node dist/index.js
+```
+
+## Quick Start
+
+```bash
+# Set up API key
+export ANTHROPIC_API_KEY=sk-ant-...
+
+# Run in interactive mode
+agent
+
+# Run a one-shot prompt
+agent "Explain the architecture of this project"
+
+# Run with explicit provider and model
+agent -p openai -m gpt-4o "Review this code"
+```
+
+## CLI Commands
+
+### Basic Usage
+
+```
+agent [options] [prompt]
+```
+
+| Option | Short | Description |
+|--------|-------|-------------|
+| `--provider <type>` | `-p` | Provider: anthropic, openai, deepseek, local, groq, xai, mistral, openrouter, gemini, moonshot |
+| `--model <name>` | `-m` | Model name (defaults per provider) |
+| `--interactive` | `-i` | Force interactive mode |
+| `--tui` | | Full-screen terminal UI |
+| `--max-turns <n>` | | Max turns (default: 40) |
+| `--max-context <n>` | | Max context tokens (default: 200000) |
+| `--continue` | | Resume most recent session |
+| `--session <id>` | | Resume specific session |
+| `--start-model` | | Start local model server |
+| `--skip-setup` | | Skip first-run setup wizard |
+
+### Subcommands
+
+#### Setup Wizard
+
+```bash
+agent setup
+```
+
+Interactive first-run configuration: API keys, default provider, model preferences.
+
+#### HTTP Server
+
+```bash
+agent serve --port 3000 --provider anthropic
+```
+
+Start the REST API server. See [API Reference](api.md) for endpoint details.
+
+#### Session Management
+
+```bash
+# List sessions for current project
+agent session list
+
+# List sessions for a specific project
+agent session list --project /path/to/project
+
+# Delete a session
+agent session delete 20260524-182358-c11f
+
+# Export session as JSON
+agent session export 20260524-182358-c11f -o ./session-backup.json
+```
+
+## Configuration
+
+### Environment Variables
+
+| Variable | Required For |
+|----------|--------------|
+| `ANTHROPIC_API_KEY` | Anthropic provider |
+| `OPENAI_API_KEY` | OpenAI provider |
+| `DEEPSEEK_API_KEY` | DeepSeek provider |
+| `GROQ_API_KEY` | Groq provider |
+| `XAI_API_KEY` | xAI provider |
+| `MISTRAL_API_KEY` | Mistral provider |
+| `OPENROUTER_API_KEY` | OpenRouter provider |
+| `GEMINI_API_KEY` / `GOOGLE_API_KEY` | Gemini provider |
+| `MOONSHOT_API_KEY` | Moonshot provider |
+| `LOG_LEVEL` | Log level: debug, info, warn, error, off (default: info) |
+
+### Config File
+
+Place `.agent/config.json` in your project directory:
+
+```json
+{
+  "provider": {
+    "type": "anthropic",
+    "model": "claude-sonnet-4-20250514"
+  },
+  "maxTurns": 40,
+  "maxContext": 200000
+}
+```
+
+### MCP Servers
+
+Configure MCP tools in `.agent/mcp.json` or `.mcp.json`:
+
+```json
+{
+  "mcpServers": {
+    "my-server": {
+      "command": "node",
+      "args": ["server.js"],
+      "env": { "KEY": "value" }
+    },
+    "remote-server": {
+      "url": "https://example.com/sse",
+      "headers": { "Authorization": "Bearer token" }
+    }
+  }
+}
+```
+
+## Custom Skills
+
+Create `.md` files with YAML frontmatter in `~/.agent/skills/` (user-level) or `.agent/skills/` (project-level). Project skills override user skills with the same name.
+
+```markdown
+---
+name: my-skill
+description: A custom skill for my workflow
+tools: read, bash, glob
+---
+
+Perform the following task based on the user's request:
+
+{{task}}
+
+Focus on accuracy and completeness.
+```
+
+Skills are hot-reloaded automatically — just save the file.
+
+## Provider Fallback
+
+Configure provider fallback chain in `.agent/config.json`:
+
+```json
+{
+  "provider": {
+    "type": "anthropic",
+    "model": "claude-sonnet-4-20250514",
+    "fallbacks": [
+      { "type": "openai", "model": "gpt-4o" },
+      { "type": "deepseek", "model": "deepseek-chat" }
+    ]
+  }
+}
+```
+
+When the primary provider fails (after retry + circuit breaker), the system automatically tries fallbacks in order.
+
+## Sub-Agent Delegation
+
+The built-in sub-agents are available as tools:
+
+| Agent | Role |
+|-------|------|
+| `code-reviewer` | Review code for bugs and style issues |
+| `security-auditor` | Audit code for security vulnerabilities |
+| `test-writer` | Write and improve tests |
+
+Invoke them by asking the agent: "Have code-reviewer review this file" or "Ask test-writer to write tests for this module."
+
+Three collaboration modes:
+- **Delegate**: Single agent handles the task
+- **Adversarial**: Two agents independently review, results are merged
+- **Parallel**: Multiple agents work simultaneously
+
+## Local Models
+
+```bash
+# Start a local model and use it
+agent --provider local --start-model --model "qwen2.5-coder:7b"
+
+# Or start local model server first, then connect
+agent --provider local
+```
+
+Supports llama.cpp, ollama, text-generation-webui, and vllm backends.
+
+## Plugins
+
+Place plugin directories in `.agent/plugins/`. Each plugin requires:
+- `manifest.json` (metadata)
+- Entry module (default: `index.ts`)
+
+Example plugin structure:
+
+```
+.agent/plugins/my-plugin/
+├── manifest.json
+└── index.ts
+```
+
+## Troubleshooting
+
+- **"No sessions found"**: Run at least one interaction first, or create a session explicitly.
+- **Provider errors**: Check `LOG_LEVEL=debug` for detailed logs. Verify API keys in environment.
+- **MCP connection failures**: Confirm the server command/URL is correct and the server is running.
+- **Tests**: Run `pnpm test` to verify system health (138+ tests).

package/package.json

@@ -0,0 +1,63 @@
+{
+  "name": "hyacinth-ai",
+  "version": "0.9.6",
+  "description": "Hyacinth 🪻 — multi-provider AI agent with tool system and context management",
+  "type": "module",
+  "main": "dist/index.js",
+  "bin": {
+    "hyacinth": "dist/index.js"
+  },
+  "keywords": [
+    "ai",
+    "agent",
+    "cli",
+    "llm",
+    "mcp",
+    "anthropic",
+    "openai",
+    "deepseek"
+  ],
+  "author": "孑遗",
+  "license": "MIT",
+  "dependencies": {
+    "@anthropic-ai/sdk": "^0.98.0",
+    "@clack/prompts": "^1.4.0",
+    "@earendil-works/pi-tui": "^0.75.4",
+    "@fastify/static": "^9.1.3",
+    "@google/genai": "^2.6.0",
+    "@larksuiteoapi/node-sdk": "^1.66.1",
+    "@modelcontextprotocol/sdk": "^1.29.0",
+    "better-sqlite3": "^12.10.0",
+    "chalk": "^5.6.2",
+    "chokidar": "^5.0.0",
+    "chrome-devtools-mcp": "^1.2.0",
+    "commander": "^14.0.3",
+    "diff": "^9.0.0",
+    "fastify": "^5.8.5",
+    "js-tiktoken": "^1.0.21",
+    "mammoth": "^1.12.0",
+    "mcp-chrome-bridge": "^1.0.31",
+    "openai": "^6.39.0",
+    "picocolors": "^1.1.1",
+    "sharp": "^0.35.0",
+    "ws": "^8.21.0",
+    "xlsx": "^0.18.5"
+  },
+  "devDependencies": {
+    "@types/better-sqlite3": "^7.6.13",
+    "@types/node": "^25.9.1",
+    "@types/ws": "^8.18.1",
+    "typescript": "^6.0.3",
+    "undici-types": "^7.24.6",
+    "vitest": "^4.1.7"
+  },
+  "scripts": {
+    "build": "tsc && node scripts/copy-prompts.cjs && node scripts/clean-tests.cjs && node scripts/clean-maps.cjs",
+    "dev": "tsc --watch",
+    "start": "node dist/index.js",
+    "start:win": "chcp 65001 > nul && node dist/index.js",
+    "test": "vitest run",
+    "test:watch": "vitest",
+    "setup:llamacpp": "powershell -ExecutionPolicy Bypass -File scripts/setup-llamacpp.ps1"
+  }
+}

package/plugins/example-greeter/index.js

@@ -0,0 +1,55 @@
+/**
+ * Example Greeter Plugin
+ *
+ * A minimal example plugin that registers a "hello" tool.
+ * This demonstrates the plugin system end-to-end.
+ */
+
+/** @type {import('../../src/plugins/types.js').PluginDefinition} */
+export default {
+  id: 'example-greeter',
+  name: 'Example Greeter',
+  description: 'Registers a hello tool and a greeting skill',
+
+  register: (api) => {
+    // ── Register a tool ───────────────────────────────────────
+    api.registerTool({
+      name: 'hello',
+      description: 'Greet the user with a friendly message',
+      inputSchema: {
+        type: 'object',
+        properties: {
+          name: {
+            type: 'string',
+            description: 'Name to greet',
+          },
+        },
+        required: [],
+      },
+      execute: async (args) => {
+        const name = args.name || 'world';
+        return `Hello, ${name}! Greetings from the example-greeter plugin.`;
+      },
+    });
+
+    api.logger.info('hello tool registered');
+
+    // ── Register a skill ──────────────────────────────────────
+    api.registerSkill({
+      name: 'example-greeter-friendly',
+      description: 'Greet the user warmly and ask about their day',
+      promptTemplate: `You are a friendly assistant. Greet {{name}} warmly and ask about their day.`,
+      relatedTools: ['hello'],
+    });
+
+    api.logger.info('example-greeter-friendly skill registered');
+  },
+
+  onActivate: async (api) => {
+    api.logger.info('Plugin activated! Config:', api.getConfig());
+  },
+
+  onDeactivate: async (api) => {
+    api.logger.info('Plugin deactivated. Goodbye!');
+  },
+};

package/plugins/example-greeter/plugin.json

@@ -0,0 +1,8 @@
+{
+  "id": "example-greeter",
+  "name": "Example Greeter",
+  "description": "A simple example plugin that registers a hello tool",
+  "entry": "./index.js",
+  "enabledByDefault": true,
+  "version": "1.0.0"
+}

package/pnpm-workspace.yaml

@@ -0,0 +1,5 @@
+ignoredBuiltDependencies:
+  - '@google/genai'
+  - protobufjs
+
+onlyBuiltDependencies: better-sqlite3