Add metadata validation script and module documentation

- Introduced a new Python script `check_agents_metadata.py` for validating agent metadata, including YAML parsing, rating ranges, and cross-references.
- Added usage instructions and exit codes for the script.
- Created a new markdown file `模块目录和功能说明.md` to outline the directory structure and functionality of the modules.
- Added a text file `说明此文件不可AI修改.txt` to specify that certain files should not be modified by AI, including important information about the `dstalk` framework and its modules.
This commit is contained in:
2026-05-31 00:00:58 +08:00
parent 3cc9ee95e4
commit f2da0f2ed4
43 changed files with 2467 additions and 800 deletions

View File

@@ -1,6 +1,13 @@
// plugin-session: 会话管理服务插件
// 提供 dstalk_session_service_t vtable 实现
// 依赖: file_io (save/load 需要文件操作)
/*
* @file session_plugin.cpp
* @brief Session plugin: conversation message history management with save/load.
* 会话插件:对话消息历史管理,支持保存/加载。
* Copyright (c) 2026 dstalk contributors. GPLv3.
*/
// plugin-session: 会话管理服务插件 / Session management service plugin
// 提供 dstalk_session_service_t vtable 实现 / Provides dstalk_session_service_t vtable implementation
// 依赖: file_io (save/load 需要文件操作) / Depends on: file_io (save/load needs file operations)
#include "dstalk/dstalk_host.h"
#include "dstalk/dstalk_types.h"
#include "dstalk/dstalk_services.h"
@@ -24,14 +31,14 @@
namespace json = boost::json;
// ============================================================
// 内部 C++ 数据结构
// 内部 C++ 数据结构 / Internal C++ data structures
// ============================================================
// W14.3: g_host / g_file_io 使用 atomic 指针,写入 acquire/release读取无锁
// W14.3: g_host / g_file_io 使用 atomic 指针,写入 acquire/release读取无锁 / g_host / g_file_io use atomic pointers, write with acquire/release, read lock-free
static std::atomic<const dstalk_host_api_t*> g_host{nullptr};
static std::atomic<const dstalk_file_io_service_t*> g_file_io{nullptr};
// 内部消息结构C++ 易用,外部暴露 C struct
// 内部消息结构C++ 易用,外部暴露 C struct / Internal message struct (C++ friendly, externally exposed as C struct)
struct InternalMessage {
std::string role;
std::string content;
@@ -39,21 +46,24 @@ struct InternalMessage {
std::string tool_calls_json;
};
// 会话历史 + 缓存 —— W14.3: mutex 保护读写
// 会话历史 + 缓存 —— W14.3: mutex 保护读写 / Session history + cache — W14.3: mutex protects read/write
static std::vector<InternalMessage> g_history;
static std::vector<dstalk_message_t> g_cached_history;
static std::mutex g_session_mutex;
// ============================================================
// Token 计数工具(内联,避免硬依赖 context 头文件)
// Token 计数工具(内联,避免硬依赖 context 头文件) / Token counting utilities (inline, avoids hard dep on context headers)
// ============================================================
// 如果字节是 ASCII (0x000x7F) 则返回 true / Returns true if the byte is ASCII (0x000x7F).
static bool is_ascii(unsigned char c) { return c < 0x80; }
// 启发式判断:如果字节起始一个 UTF-8 CJK 统一表意文字 (0xE40xE9) 则返回 true / Heuristic: returns true if the byte starts a CJK Unified Ideograph in UTF-8 (0xE40xE9).
static bool starts_cjk(unsigned char c) {
return c >= 0xE4 && c <= 0xE9;
}
// 使用启发式 UTF-8 字节计数估算单条消息的 token 数 / Estimate token count for a single message using heuristic UTF-8 byte counting.
static size_t count_tokens_one(const std::string& text) {
size_t ascii_chars = 0;
size_t chinese_chars = 0;
@@ -85,9 +95,10 @@ static size_t count_tokens_one(const std::string& text) {
}
size_t content_tokens = (ascii_chars / 4) + (chinese_chars / 2) + (other_chars / 3);
return content_tokens + 4; // +4 per message overhead
return content_tokens + 4; // +4 每条消息开销 / +4 per message overhead
}
// 估算所有消息的总 token 数 / Estimate total token count across all messages.
static size_t count_tokens_all(const std::vector<InternalMessage>& msgs) {
size_t total = 0;
for (const auto& m : msgs) {
@@ -97,13 +108,15 @@ static size_t count_tokens_all(const std::vector<InternalMessage>& msgs) {
}
// ============================================================
// 辅助:刷新 C 缓存数组(调用方需持有 g_session_mutex
// 辅助:刷新 C 缓存数组(调用方需持有 g_session_mutex / Helper: rebuild C cached array (caller must hold g_session_mutex)
// ============================================================
// 从内部消息 vector 重建 C 兼容的缓存历史数组。调用方必须持有 g_session_mutex / Rebuild the C-compatible cached history array from the internal message vector.
// Caller must hold g_session_mutex.
static void rebuild_cached_history_locked() {
const dstalk_host_api_t* host = g_host.load(std::memory_order_acquire);
// 释放旧的字符串
// 释放旧的字符串 / Free old strings
for (auto& m : g_cached_history) {
if (m.role) { host->free(const_cast<char*>(m.role)); }
if (m.content) { host->free(const_cast<char*>(m.content)); }
@@ -112,7 +125,7 @@ static void rebuild_cached_history_locked() {
}
g_cached_history.clear();
// 重建
// 重建 / Rebuild
g_cached_history.reserve(g_history.size());
for (const auto& im : g_history) {
dstalk_message_t cm;
@@ -125,9 +138,10 @@ static void rebuild_cached_history_locked() {
}
// ============================================================
// Session 服务 vtable 实现 (W14.3: try/catch + mutex)
// Session 服务 vtable 实现 (W14.3: try/catch + mutex) / Session service vtable implementation (W14.3: try/catch + mutex)
// ============================================================
// 向对话历史追加一条消息 / Append a message to the conversation history.
static void session_add(const dstalk_message_t* msg) {
try {
if (!msg) return;
@@ -148,11 +162,13 @@ static void session_add(const dstalk_message_t* msg) {
}
}
// 清空对话历史中的所有消息 / Clear all messages from the conversation history.
static void session_clear() {
std::lock_guard<std::mutex> lock(g_session_mutex);
g_history.clear();
}
// 将当前对话历史序列化为 JSON 行文件并保存到 path / Serialize the current conversation history to a JSON lines file at `path`.
static int session_save(const char* path) {
try {
if (!path) return -1;
@@ -187,6 +203,7 @@ static int session_save(const char* path) {
}
}
// 从 JSON 行文件中加载对话历史,替换当前历史 / Load conversation history from a JSON lines file at `path`, replacing current history.
static int session_load(const char* path) {
try {
if (!path) return -1;
@@ -246,6 +263,7 @@ static int session_load(const char* path) {
}
}
// 返回指向缓存 C 消息数组的指针,并将 *out_count 设置为数组大小 / Return a pointer to the cached C-array of messages and set *out_count to its size.
static const dstalk_message_t* session_history(int* out_count) {
try {
std::lock_guard<std::mutex> lock(g_session_mutex);
@@ -265,6 +283,7 @@ static const dstalk_message_t* session_history(int* out_count) {
}
}
// 返回当前对话历史的估算 token 数 / Return the estimated token count for the current conversation history.
static int session_token_count() {
try {
std::lock_guard<std::mutex> lock(g_session_mutex);
@@ -290,11 +309,12 @@ static dstalk_session_service_t g_session_service = {
};
// ============================================================
// W20.6: 默认会话保存路径(平台标准目录)
// W20.6: 默认会话保存路径(平台标准目录) / Default session save path (platform standard directory)
// ============================================================
// 返回平台特定的默认会话保存路径,根据需要创建目录 / Return the platform-specific default session save path, creating directories as needed.
static std::string get_default_session_path() {
// W22.5: static 缓存 + mkdir 保障 + 失败 fallback 到当前目录
// W22.5: static 缓存 + mkdir 保障 + 失败 fallback 到当前目录 / static cache + mkdir guarantee + fallback to current dir on failure
static std::string cached_path = []() -> std::string {
#ifdef _WIN32
char* buf = nullptr;
@@ -323,14 +343,17 @@ static std::string get_default_session_path() {
}
// ============================================================
// 插件生命周期
// 插件生命周期 / Plugin lifecycle
// ============================================================
// 插件初始化:保存主机指针,查询 file_io 依赖,注册 session 服务,
// 并从默认路径自动加载已有会话 / Plugin init: store host pointer, query file_io dependency, register session service,
// and auto-load any existing session from the default path.
static int on_init(const dstalk_host_api_t* host) {
try {
g_host.store(host, std::memory_order_release);
// 查询依赖服务: file_io
// 查询依赖服务: file_io / Query dependency service: file_io
void* raw = host->query_service("file_io", 1);
if (!raw) {
host->log(DSTALK_LOG_ERROR, "[plugin-session] required service 'file_io' not found");
@@ -338,11 +361,11 @@ static int on_init(const dstalk_host_api_t* host) {
}
g_file_io.store(static_cast<const dstalk_file_io_service_t*>(raw), std::memory_order_release);
// 注册自身服务
// 注册自身服务 / Register own service
int ret = host->register_service("session", 1, &g_session_service);
if (ret != 0) return ret;
// W20.6: 从默认路径恢复会话(文件不存在则静默失败)
// W20.6: 从默认路径恢复会话(文件不存在则静默失败) / Restore session from default path (silent fail if file missing)
session_load(get_default_session_path().c_str());
return 0;
@@ -357,10 +380,13 @@ static int on_init(const dstalk_host_api_t* host) {
}
}
// 插件关闭:自动保存会话到默认路径,失败时回退到当前目录,
// 然后释放缓存历史和清空状态 / Plugin shutdown: auto-save session to default path, fallback to current dir on failure,
// then release cached history and clear state.
static void on_shutdown() {
try {
// W20.6: 清空前自动保存到默认路径
// W21.4: 失败告警 + 当前目录 fallback
// W20.6: 清空前自动保存到默认路径 / Auto-save to default path before clearing
// W21.4: 失败告警 + 当前目录 fallback / Failure warning + current dir fallback
int ret = session_save(get_default_session_path().c_str());
if (ret != 0) {
const dstalk_host_api_t* h = g_host.load(std::memory_order_acquire);
@@ -389,7 +415,7 @@ static void on_shutdown() {
static dstalk_plugin_info_t g_info = {
"session",
"1.0.0",
"Session management plugin with save/load support",
"Session management plugin with save/load support / 支持保存/加载的会话管理插件",
DSTALK_API_VERSION,
{"file_io", nullptr, nullptr, nullptr, nullptr, nullptr, nullptr, nullptr},
on_init,
@@ -397,6 +423,7 @@ static dstalk_plugin_info_t g_info = {
nullptr
};
// 必须入口点:返回插件描述符给主机 / Mandatory entry point: returns the plugin descriptor to the host.
extern "C" DSTALK_PLUGIN_EXPORT dstalk_plugin_info_t* dstalk_plugin_init(void) {
return &g_info;
}