Skip to content

v3.1.7

Choose a tag to compare

@hellowind777 hellowind777 released this 28 Jun 12:54
· 9 commits to main since this release
b7eeb85

中文

新特性

  • 新增 ~ask 命令:交互式需求澄清入口,一问一答厘清目标、方向、范围与约束,纯对话不创建文件。合并原 ~idea(方向探索)和 ~office(价值评估/范围收缩),提供统一入口
  • ~plan 需求澄清阶段默认采用交互澄清模式(一问一答、选择题优先),已有代码库时先静默读代码再以选择题逐条确认,原"假设模式"改为快速通道(用户说"直接做,不用问我"时启用)
  • ~prd 项目定位和维度探索阶段改为一问一答交互澄清,维度探索不再"AI 先给推荐让用户审查",改为逐条提问带推荐选项
  • ~build 无方案包且任务 T2+ 时自动重定向到 ~plan,避免需求不清时直接写代码
  • tasks.md 模板完成标准改为"可独立验证的布尔条件"——任何人读完就能判断任务是否完成
  • qa-review 新增任务完成标准逐条核对步骤,验证不依赖主会话上下文

问题修复

  • 修复 ~idea 兼容别名在脚本路由层失效问题(COMMAND_ALIASES 缺少 idea: 'ask' 映射)
  • 统一所有命令 SKILL.md 中 AI 指令语气为第二人称"你"(原 ~ask 和 ~prd 中混用"AI"第三人称)
  • 修复 guard 测试中断言与当前 guard.mjs 行为不一致的预存问题(高危命令由 deny 改为 warn 后测试未同步)
  • 统一"可独立验证"语序(原"独立可验证"共 2 处不一致)
  • bootstrap.md / bootstrap-lite.md 中 "SKILL" 加反引号,措辞"温和偏差拖小"改为"惯性偏差拖累"
  • bootstrap-lite.md 补充命令执行指引,避免精简模式下代理不清楚后续流程
  • README_CN.md 首页长句优化为更自然的表达

文档

  • 全面更新 README 中英文版本中的命令表、别名和描述

杂项

  • zeroSideEffect 标记添加注释说明设计意图

变更记录

完整更新记录:v3.1.5 beta (891bab2) ... v3.1.7 beta

  • feat 新增 ~ask 交互式需求澄清命令,合并 ~idea + ~office 功能
  • feat ~plan / ~prd / ~build 全面采用交互澄清模式
  • feat 任务完成标准改为"可独立验证的布尔条件"
  • fix COMMAND_ALIASES 补充 idea: 'ask' 映射
  • fix 统一所有命令 SKILL.md 语气为第二人称"你"
  • fix 修复 guard 测试预存断言不一致问题
  • fix 统一措辞、修复语言质量问题
  • release 同步 v3.1.7 beta 版本号

English

New Features

  • New ~ask command: interactive requirements clarification via structured Q&A. Covers direction, value judgment, scope, and premise challenging — pure conversation, no files created. Merges former ~idea (exploration) and ~office (value/scope review) into a single entry point
  • ~plan now defaults to interactive clarification mode (one question at a time, multiple-choice preferred). Existing codebases get silent code reading first, then choice-based confirmation. Former "assumption mode" becomes a fast-track option ("just do it, don't ask me")
  • ~prd project positioning and dimension exploration now use interactive Q&A. Dimension exploration no longer starts with "AI recommends first, user reviews" — it now asks one question at a time with recommended options
  • ~build now redirects to ~plan when no plan package exists and task is T2+, preventing premature coding with unclear requirements
  • tasks.md template completion criteria updated to "independently verifiable boolean conditions" — anyone can judge task completion by reading the criteria alone
  • qa-review now includes per-task completion criteria verification, independent of main session context

Bug Fixes

  • Fixed ~idea compatibility alias broken at script routing layer (COMMAND_ALIASES missing idea: 'ask' mapping)
  • Unified AI directive voice to second-person "you" across all command SKILL.md files (previously mixed "AI" third-person in ~ask and ~prd)
  • Fixed stale guard test assertions mismatched with current guard.mjs behavior (HIGH_RISK commands changed from deny to warn)
  • Unified word order for "independently verifiable" across all files (2 instances were inconsistent)
  • Added backticks to "SKILL" in bootstrap.md / bootstrap-lite.md; improved phrasing
  • Added command execution guidance to bootstrap-lite.md for clarity in lightweight mode
  • Improved long-sentence readability in README_CN.md

Documentation

  • Comprehensively updated README (EN/CN) command table, aliases, and descriptions

Chores

  • Added explanatory comment to zeroSideEffect flag documenting design intent

Changelog

Full Changelog: v3.1.5 beta (891bab2) ... v3.1.7 beta

  • feat add ~ask interactive clarification command, merge ~idea + ~office
  • feat adopt interactive clarification mode in ~plan / ~prd / ~build
  • feat update task completion criteria to independently verifiable boolean conditions
  • fix add idea: 'ask' to COMMAND_ALIASES
  • fix unify all command SKILL.md voice to second-person
  • fix sync stale guard test assertions
  • fix unify wording and fix language quality issues
  • release bump to v3.1.7 beta