Releases
v0.3.0
ZX-Code v0.3.0 - 技能进化引擎 + 定时Agent + 用户画像 + 执行轨迹 + 脚本沙箱 + 历史搜索
Compare
Sorry, something went wrong.
No results found
Bug Fixes (2026-07-13 重新构建)
Agent 系统修复
autoAccept 绕过 deny 规则 — 修复 autoAccept=true 时跳过整个权限检查块导致 deny 规则失效的安全漏洞,现在 deny 规则在 autoAccept 下始终生效
工具参数解析失败仍执行 — 修复 JSON.parse 失败后 parsedArgs 保持空对象导致工具用空参数执行的问题,现在参数解析失败时返回 is_error 并跳过执行
pendingQuestions 资源泄漏 — 修复 chat:stop 时未清理 onQuestion Promise 导致 10 分钟超时和资源泄漏的问题,新增 cleanupPendingQuestions 在 stop/forceReset 时调用
UI 修复
Toggle 开关白点卡死 — 修复 framer-motion layout 属性与手动 style.transform 冲突导致开关白点关闭再打开后锁在左边的问题,改用 animate.x 管理位置
自动接受工具调用开关失效 — 修复 Toggle 视觉状态与实际 checked 值不同步导致自动接受工具调用开关显示为开但实际为关的问题
测试
新增 4 个 Toggle 测试 + 3 个 engine.handleToolCalls 测试 + 2 个 chat.ipc.questions 测试
全部 610 个 renderer 测试通过
ZX-Code v0.3.0 更新日志 ## 新增功能 ### 技能进化引擎 - GEPA 自我改进 — 基于 Reflective Evolution 理念,自动生成技能变体并用 LLM-as-Judge 评分(adherence / correctness / conciseness) - 语义保留检查 — 进化变体需通过语义偏差检测,确保核心意图不变 - 提升阈值部署 — 仅当变体分数 ≥ 基线 × 1.1 时才部署新版本,避免退化 - 技能自动创建 — Agent 完成复杂任务后自动抽取经验沉淀为可复用技能(skill_create 工具) - 版本管理 — 每次进化保留完整版本历史,可回滚到任意版本 ### 定时 Agent (Cron) - Cron 表达式调度 — 标准 5 字段 cron 格式,如 0 9 * * * 每天 9 点执行 - 自主创建 — Agent 通过 cron_manage 工具创建/列出/删除/切换定时任务 - 安全约束 — allowWriteTools=false 时自动过滤写入类工具 - 执行记录 — 记录每次运行结果、运行时长、摘要 ### 用户画像学习 - 自动抽取 — 对话结束后异步抽取用户画像(技术栈、编码风格、工作模式、沟通偏好、专业水平、语言偏好) - 置信度评分 — 每条画像条目携带 confidence 值 - System Prompt 注入 — 自动将用户画像摘要注入 system message,让 Agent 更懂你 ### 执行轨迹追踪 - 全量记录 — 记录每次 Agent 执行的完整轨迹:迭代轮次、工具调用、参数摘要、结果摘要、耗时、成功/失败 - 统计分析 — 轨迹统计:总轨迹数、总工具调用数、平均耗时、成功率、Top 工具排行 - 多维查询 — 支持按对话 ID、工具名、成功/失败、时间范围查询 - 技能进化数据源 — 执行轨迹作为技能进化评估数据集,实现闭环优化 ### 脚本沙箱 (run_script) - 隔离执行 — 在安全沙箱中执行 JavaScript 代码,可通过 tools.<name>(args) RPC 调用所有已注册工具 - 安全隔离 — 沙箱中不提供 require / process / Buffer 等危险全局对象 - 批量操作 — 适用于批量文件操作、组合工具调用、数据清洗 ### 历史全文搜索 - FTS5 全文索引 — 对话历史基于 SQLite FTS5,支持消息级和对话级搜索 - 文件搜索增强 — 支持文件名 / 内容 / 全部模式,支持正则、大小写、全词匹配 ### 新增内置工具 - cron_manage — 管理定时 Agent 任务(创建/列出/删除/切换) - run_script — 在隔离沙箱中执行 JavaScript 脚本 - skill_create — 从当前任务经验中创建可复用技能 --- ## Changelog (English) ### New Features #### Skill Evolution Engine - GEPA self-improvement — Auto-generates skill variants and scores them with LLM-as-Judge (adherence / correctness / conciseness) - Semantic preservation check — Variants must pass semantic deviation detection - Improvement threshold deployment — New versions deployed only when score ≥ baseline × 1.1 - Auto skill creation — Agent extracts reusable skills from complex task experiences (skill_create tool) - Version management — Full version history with rollback support #### Scheduled Agent (Cron) - Cron expression scheduling — Standard 5-field cron format - Self-managed — Agent creates/lists/deletes/toggles scheduled tasks via cron_manage tool - Safety constraints — allowWriteTools=false auto-filters write tools - Execution records — Logs each run's result, duration, and summary #### User Profile Learning - Auto-extraction — Async extraction of user profile (tech stack, coding style, work patterns, communication preferences, expertise level, language preference) - Confidence scoring — Each profile entry carries a confidence value - System prompt injection — Automatically injects user profile summary into system message #### Execution Trace Tracking - Full recording — Records complete traces: iteration rounds, tool calls, parameter summaries, result summaries, duration, success/failure - Analytics — Trace stats: total traces, total tool calls, average duration, success rate, top tools - Multi-dimensional queries — Query by conversation ID, tool name, success/failure, time range - Skill evolution data source — Traces serve as evaluation datasets for skill evolution #### Script Sandbox (run_script) - Isolated execution — Execute JavaScript in a secure sandbox with RPC access to all registered tools - Security isolation — No dangerous globals (require / process / Buffer) - Batch operations — Suitable for batch file operations, combined tool calls, data cleaning #### Full-text History Search - FTS5 full-text index — Conversation history indexed via SQLite FTS5 - Enhanced file search — Supports filename / content / all modes with regex, case-sensitive, whole-word matching #### New Built-in Tools - cron_manage — Manage scheduled Agent tasks (create/list/delete/toggle) - run_script — Execute JavaScript scripts in an isolated sandbox - skill_create — Create reusable skills from current task experiences
You can’t perform that action at this time.