✨ feat/gpio_tools: 更新gpio tool，AI可以通过tool直接调用指定IO

README.md

@@ -4,6 +4,8 @@
 
 <div align="center">
 
+![logo](assets/logo.png)
+
 **Decouple LLM, Tools, Agent, and Channels—then pack them onto a single ESP32-S3.**
 
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE) ![ESP32-S3](https://img.shields.io/badge/MCU-ESP32--S3-ff6a00) ![ESP-IDF](https://img.shields.io/badge/ESP--IDF-v5.5.2-00979D) ![LLM](https://img.shields.io/badge/LLM-Qwen%20via%20DashScope-0f766e) ![Channel](https://img.shields.io/badge/Channel-Feishu%20%7C%20WebSocket%20%7C%20QQBot-2563eb) ![Search](https://img.shields.io/badge/Search-Tavily-111827)
@@ -18,7 +20,7 @@
 
 This project draws on the ideas and direction of:
 
-- [OpenClaw](https://github.com/OpenClawAI/OpenClaw)
+- [OpenClaw](https://github.com/openclaw/openclaw)
 - [MimiClaw](https://github.com/memovai/mimiclaw)
 
 EmbedClaw keeps the goal of running a full AI Agent on low-power hardware but focuses the architecture on **decoupling LLM, Tools, Agent, and Channels**.  
@@ -69,10 +71,10 @@ It’s a working “embedded Agent base” you can extend.
 | Chat Channel | Feishu, WebSocket, QQBot | Feishu long connection, local WebSocket chat, official QQBot gateway |
 | Agent | ReAct tool loop | Model can call tools, read results, then continue |
 | Long-term memory | `/spiffs/memory/MEMORY.md` | User profile, preferences, stable facts |
-| Short-term memory | `/spiffs/session/se_<hash>.jsonl` | Recent conversation for current session |
+| Short-term memory | `/spiffs/session/se_<hash>.jsonl` | Conversation history including tool call traces |
 | Daily notes | `/spiffs/memory/<YYYY-MM-DD>.md` | Recent events and daily context |
-| Skills | Built-in + SPIFFS | Task instructions as Markdown |
-| Tools | Files, time, search, cron | Exposed to LLM via JSON schema |
+| Skills | SPIFFS pre-installed + runtime | Task instructions as Markdown |
+| Tools | Files, time, search, cron, GPIO | Exposed to LLM via JSON schema |
 
 ### Registered tools
 
@@ -87,10 +89,11 @@ It’s a working “embedded Agent base” you can extend.
 | `cron_add` | Add periodic or one-shot scheduled tasks |
 | `cron_list` | List scheduled tasks |
 | `cron_remove` | Remove scheduled tasks |
+| `gpio_control` | Control ESP32 GPIO pins (on, off, set, toggle, get) |
 
-### Built-in skills
+### Pre-installed skills
 
-These are installed at startup:
+These are pre-installed as Markdown files in `spiffs_data/skills/` and deployed with the SPIFFS image:
 
 - `weather`
 - `daily-briefing`
@@ -115,6 +118,7 @@ flowchart LR
     T --> S2[File Tools]
     T --> S3[Time Tool]
     T --> S4[Cron Tool]
+    T --> S5[GPIO Tool]
     A --> M1[Session Memory]
     A --> M2[Long-term Memory]
     A --> K[Skill Loader]
@@ -137,6 +141,9 @@ flowchart LR
 │   ├── embed_claw.c             # System startup entry
 │   └── ec_config_internal.h     # Built-in defaults; local overrides live in main/ec_config.h
 ├── spiffs_data/                 # Default SPIFFS image content
+│   ├── config/                  # SOUL.md, USER.md
+│   ├── memory/                  # MEMORY.md
+│   └── skills/                  # Pre-installed skill files
 └── scripts/                     # WebSocket test script and test-app helpers
 ```
 
@@ -540,7 +547,6 @@ Skills are Markdown task descriptions, not code. You can:
 
 - Write them at runtime via tools to `/spiffs/skills/<name>.md`
 - Or put default skills in `spiffs_data/skills/` so they’re in the SPIFFS image
-- Or add built-in skills in `ec_skill_loader.c`
 
 Suggested format:
 

README_ZH.md

@@ -4,6 +4,8 @@
 
 <div align="center">
 
+![logo](assets/logo.png)
+
 **把 LLM、Tools、Agent、Channel 彻底拆开，再把它们装进一块 ESP32-S3。**
 
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE) ![ESP32-S3](https://img.shields.io/badge/MCU-ESP32--S3-ff6a00) ![ESP-IDF](https://img.shields.io/badge/ESP--IDF-v5.5.2-00979D) ![LLM](https://img.shields.io/badge/LLM-Qwen%20via%20DashScope-0f766e) ![Channel](https://img.shields.io/badge/Channel-Feishu%20%7C%20WebSocket%20%7C%20QQBot-2563eb) ![Search](https://img.shields.io/badge/Search-Tavily-111827)
@@ -17,7 +19,7 @@
 
 本仓库在理念和方向上参考了以下优秀项目：
 
-- [OpenClaw](https://github.com/OpenClawAI/OpenClaw)
+- [OpenClaw](https://github.com/openclaw/openclaw)
 - [MimiClaw](https://github.com/memovai/mimiclaw)
 
 EmbedClaw 延续了“在低功耗硬件上运行完整 AI Agent”的思路，但把架构重点放在了 **LLM、Tools、Agent、Channel 的解耦** 上。  
@@ -68,9 +70,9 @@ EmbedClaw 延续了“在低功耗硬件上运行完整 AI Agent”的思路，
 | Chat Channel | Feishu、WebSocket、QQBot | 飞书长连接、本地 WebSocket 对话、官方 QQBot gateway |
 | Agent | ReAct Tool Loop | 支持模型调用工具、再读工具结果、再继续推理 |
 | 长期记忆 | `/spiffs/memory/MEMORY.md` | 用户画像、长期偏好、稳定事实 |
-| 短期记忆 | `/spiffs/session/se_<hash>.jsonl` | 最近对话历史，供当前会话上下文使用 |
+| 短期记忆 | `/spiffs/session/se_<hash>.jsonl` | 对话历史（含完整工具调用链），供当前会话上下文使用 |
 | 每日笔记 | `/spiffs/memory/<YYYY-MM-DD>.md` | 记录近期事件与每日上下文 |
-| Skills | 内置 + SPIFFS 动态技能 | 任务指令可持久化为 Markdown |
+| Skills | SPIFFS 预置 + 运行时动态添加 | 任务指令可持久化为 Markdown |
 | Tools | 文件、时间、搜索、定时任务 | 通过统一 JSON Schema 暴露给 LLM |
 
 ### 已注册 Tools
@@ -86,10 +88,11 @@ EmbedClaw 延续了“在低功耗硬件上运行完整 AI Agent”的思路，
 | `cron_add` | 创建周期任务或单次任务 |
 | `cron_list` | 查看当前定时任务 |
 | `cron_remove` | 删除定时任务 |
+| `gpio_control` | 控制 ESP32 GPIO 引脚（on/off/set/toggle/get） |
 
-### 已启用 Skills
+### 预置 Skills
 
-系统启动时会自动安装内置 Skills：
+以下技能以 Markdown 文件形式预置在 `spiffs_data/skills/` 中，随 SPIFFS 镜像一同烧录：
 
 - `weather`
 - `daily-briefing`
@@ -114,6 +117,7 @@ flowchart LR
     T --> S2[File Tools]
     T --> S3[Time Tool]
     T --> S4[Cron Tool]
+    T --> S5[GPIO Tool]
     A --> M1[Session Memory]
     A --> M2[Long-term Memory]
     A --> K[Skill Loader]
@@ -136,6 +140,9 @@ flowchart LR
 │   ├── embed_claw.c              # 系统统一启动入口
 │   └── ec_config_internal.h      # 仓库内置默认配置，项目覆盖放在 main/ec_config.h
 ├── spiffs_data/                  # 默认写入 SPIFFS 的系统文件
+│   ├── config/                   # SOUL.md、USER.md
+│   ├── memory/                   # MEMORY.md
+│   └── skills/                   # 预置 Skill 文件
 └── scripts/                      # WebSocket 测试脚本与测试构建辅助脚本
 ```
 
@@ -569,7 +576,6 @@ Skill 是最轻量的扩展方式。它不是代码，而是任务说明书。
 
 - 在运行时通过 Tool 把 Skill 写入 `/spiffs/skills/<name>.md`
 - 或者把默认 Skill 放进 `spiffs_data/skills/`，随 SPIFFS 镜像一同烧录
-- 或者在 `ec_skill_loader.c` 中增加新的内置 Skill
 
 建议格式：
 

TODO.md

@@ -17,6 +17,7 @@
 - [x] Web search
 - [ ] Script interpreter
 - [ ] Peripherals (GPIO, I2C, etc.)
+    - [X] GPIO
 - [ ] Camera
 
 ## Channel integration
@@ -33,3 +34,5 @@
 - [x] test and ci/cd
 - [ ] Support more filesystems (beyond SPIFFS)
 - [ ] SD card support
+- [ ] Support ESP32C3
+

TODO_ZH.md

@@ -17,6 +17,7 @@
 - [x] 对接web搜索
 - [ ] 对接脚本语言解释器
 - [ ] 对接外设
+    - [X] 对接GPIO
 - [ ] 对接摄像头
 
 ## Channel对接
@@ -33,3 +34,4 @@
 - [x] 测试和ci/cd
 - [ ] 更改SPIFFS为更多的文件系统
 - [ ] 支持 SD card
+- [ ] 支持 ESP32C3

components/embed_claw/CMakeLists.txt

@@ -8,6 +8,7 @@ idf_component_register(
     INCLUDE_DIRS
         "."
     REQUIRES
+        driver
         esp_http_client
         esp_http_server
         esp_netif

components/embed_claw/core/ec_agent.c

@@ -41,21 +41,15 @@
 #define EC_AGENT_CHANNEL_DELIVERY_TASK_STACK (8 * 1024)
 #define EC_AGENT_CHANNEL_DELIVERY_TASK_PRIO  5
 
-#define EC_AGENT_SYSTEM_PROMPT_STR \
+#define EC_AGENT_SYSTEM_PROMPT_HEAD \
         "You are EmbedClaw, a helpful and concise AI assistant running on an ESP32 device.\n"\
         "You communicate via Feishu and WebSocket.\n"\
         "Reply briefly to short messages (e.g. 你好, 在吗, 谢谢).\n"\
         "# Tools\n"\
-        "- web_search: search current information.\n"\
-        "- get_current_time: get date/time.\n"\
-        "- read_file: read /spiffs/ files (path must start with " EC_FS_BASE "/).\n"\
-        "- write_file: Write file.\n"\
-        "- edit_file: edit file.\n"\
-        "- list_dir: list files.\n"\
-        "- cron_add: schedule task.\n"\
-        "- cron_list: list tasks.\n"\
-        "- cron_remove: remove task.\n\n"\
-        "When using cron_add to reply later in the same conversation, reuse the current channel, chat_type, and chat_id.\n\n"\
+        "Available tools:\n"
+
+#define EC_AGENT_SYSTEM_PROMPT_TAIL \
+        "\nWhen using cron_add to reply later in the same conversation, reuse the current channel, chat_type, and chat_id.\n\n"\
         "Use tools when needed. Provide your final answer as text after using tools.\n\n"\
         "## Memory\n"\
         "You have persistent memory stored on local flash:\n"\
@@ -328,19 +322,34 @@ static esp_err_t context_build_system_prompt(char *buf, size_t size)
     size_t off = 0;
     size_t cap = size - 1;
 
-    off += snprintf(buf + off, size - off, EC_AGENT_SYSTEM_PROMPT_STR);
+    off += snprintf(buf + off, size - off, EC_AGENT_SYSTEM_PROMPT_HEAD);
     if (off > cap) {
         off = cap;
     }
 
-    /* Bootstrap files */
-    off = append_file(buf, size, off, EC_SOUL_FILE, "Personality");
-    off = append_file(buf, size, off, EC_USER_FILE, "User Info");
-
     scratch = calloc(1, EC_AGENT_PROMPT_SCRATCH_SIZE);
     if (!scratch) {
         ESP_LOGW(TAG, "Skipping optional prompt sections: out of memory");
     } else {
+        size_t tools_len = ec_tools_build_summary(scratch, EC_AGENT_PROMPT_SCRATCH_SIZE);
+        if (off < cap && tools_len > 0) {
+            off += snprintf(buf + off, size - off, "%s", scratch);
+            if (off > cap) {
+                off = cap;
+            }
+        }
+    }
+
+    off += snprintf(buf + off, size - off, EC_AGENT_SYSTEM_PROMPT_TAIL);
+    if (off > cap) {
+        off = cap;
+    }
+
+    /* Bootstrap files */
+    off = append_file(buf, size, off, EC_SOUL_FILE, "Personality");
+    off = append_file(buf, size, off, EC_USER_FILE, "User Info");
+
+    if (scratch) {
         if (off < cap && ec_memory_read_long_term(scratch, EC_AGENT_PROMPT_SCRATCH_SIZE) == ESP_OK && scratch[0]) {
             off += snprintf(buf + off, size - off, "\n## Long-term Memory\n\n%s\n", scratch);
             if (off > cap) {
@@ -377,7 +386,6 @@ static esp_err_t context_build_system_prompt(char *buf, size_t size)
     }
 
     ESP_LOGI(TAG, "System prompt built: %d bytes", (int)off);
-    ESP_LOGI(TAG, "prompt:%s", buf);
 
     return ESP_OK;
 }
@@ -455,6 +463,8 @@ static void agent_loop_task(void *arg)
         if (!messages) {
             messages = cJSON_CreateArray();
         }
+        int history_count = cJSON_GetArraySize(messages);
+
         cJSON *user_msg = cJSON_CreateObject();
         cJSON_AddStringToObject(user_msg, "role", "user");
         cJSON_AddStringToObject(user_msg, "content", msg.content);
@@ -488,22 +498,18 @@ static void agent_loop_task(void *arg)
             }
 
             if (!resp.tool_use) {
-                // 正常对话结束，保存最终文本并退出循环
                 if (resp.text && resp.text_len > 0) {
                     final_text = strdup(resp.text);
                 }
                 ec_llm_response_free(&resp);
                 break;
             }
 
-
-            // 构建助手消息，包含文本回复和工具调用信息，供工具执行结果使用
             cJSON *asst_msg = cJSON_CreateObject();
             cJSON_AddStringToObject(asst_msg, "role", "assistant");
             cJSON_AddItemToObject(asst_msg, "content", build_assistant_content(&resp));
             cJSON_AddItemToArray(messages, asst_msg);
 
-            // 执行工具并将结果追加到消息数组中，供下一轮 LLM 调用使用
             cJSON *tool_results = build_tool_results(&resp, &msg, tool_output, EC_AGENT_TOOL_OUTPUT_SIZE);
             cJSON *result_msg = cJSON_CreateObject();
             cJSON_AddStringToObject(result_msg, "role", "user");
@@ -514,14 +520,19 @@ static void agent_loop_task(void *arg)
         }
 
         if (final_text && final_text[0]) {
-            // 保存用户消息和助手回复到会话历史中
-            esp_err_t save_user = ec_session_append(session_key, "user", msg.content);
-            esp_err_t save_asst = ec_session_append(session_key, "assistant", final_text);
-            if (save_user != ESP_OK || save_asst != ESP_OK) {
-                ESP_LOGW(TAG, "Session save failed for %s:%s:%s (user=%s, assistant=%s)",
-                         msg.channel, msg.chat_type, msg.chat_id,
-                         esp_err_to_name(save_user),
-                         esp_err_to_name(save_asst));
+            int total = cJSON_GetArraySize(messages);
+            bool save_ok = true;
+            for (int k = history_count; k < total; k++) {
+                if (ec_session_append_msg(session_key, cJSON_GetArrayItem(messages, k)) != ESP_OK) {
+                    save_ok = false;
+                }
+            }
+            if (ec_session_append(session_key, "assistant", final_text) != ESP_OK) {
+                save_ok = false;
+            }
+            if (!save_ok) {
+                ESP_LOGW(TAG, "Session save failed for %s:%s:%s",
+                         msg.channel, msg.chat_type, msg.chat_id);
             } else {
                 ESP_LOGI(TAG, "Session saved for %s:%s:%s", msg.channel, msg.chat_type, msg.chat_id);
             }