Waza V3.28.0

Broader agent installs, easier updates, and stricter release gates.

Changelog

1. Install: Adds Antigravity, Antigravity CLI, OpenCode, and Pi install paths to README and package metadata.
2. Release: Publishes @tw93/waza from GitHub releases and verifies the npm version plus latest dist-tag before the workflow passes.
3. Installer: Moves README curl commands to latest release assets while keeping installer scripts pinned to the release tag they ship with.
4. Verifier: Adds checks for npm release workflow coverage, installer assets, stale resolver references, package stragglers, and default-deny packaging behavior.
5. Check and Hunt: Strengthens release-gate review, instrument-first debugging, native freeze triage, and recurring bug testability rules.
6. Write: Adds product localization review, long-form article structure cleanup, and sharper English rules for diagrams, lists, and recap cuts.
7. Design and Health: Adds multilingual typography guidance, responsive screenshot checks, safer agent supply-chain audit rules, and concentrated fix-chain detection.
8. Read and Statusline: Adds fetch timeouts and safer temp-file handling, improves Feishu parsing, and shows live rate-limit values instead of stale high-water marks.

更新日志

1. Install: 在 README 和 package metadata 中加入 Antigravity、Antigravity CLI、OpenCode 和 Pi 安装路径。
2. Release: GitHub release 会发布 @tw93/waza，并在 workflow 通过前回读 npm 版本和 latest dist-tag。
3. Installer: README 的 curl 命令改用 latest release asset，同时 installer 脚本内部仍固定到随 release 发布的 tag。
4. Verifier: 新增 npm release workflow、installer assets、stale resolver、package straggler 和 default-deny packaging 的检查。
5. Check and Hunt: 加强 release gate review、instrument-first 排障、native freeze 分析，以及反复出现 bug 的 testability 规则。
6. Write: 新增产品本地化审阅、长文结构清理，以及面向图表、列表和复述删减的英文规则。
7. Design and Health: 加入多语言字体、响应式截图检查、更安全的 agent supply-chain 审计，以及集中 fix 链路检测。
8. Read and Statusline: 增加 fetch timeout 和更安全的临时文件处理，改进飞书解析，并显示实时 rate-limit 值而不是旧 high-water mark。

Update: npx skills add tw93/Waza@latest · Claude Desktop · ⭐ tw93/Waza

Waza

Clearer routing, safer release checks, and cleaner project memory distillation.

Changelog

1. Routing: Adds a shared Waza routing rule and installer path so hosts that do not route from skill descriptions can still pick the right skill.
2. Verifier: Adds checks that keep routing rules, quoted triggers, and anti-pattern rows consistent, while catching stale private-surface wording.
3. Check: Improves issue triage, shipped-state answers, generated-file checks, contributor-branch edits, verifier-failure triage, cleanup safety, migration scope, and dead-code proof.
4. Read: Finds helper scripts correctly from source repos, single-skill installs, and packaged dispatchers, while removing the old save-path reference.
5. Design and Hunt: Adds quieter product-boundary rules, spacing and motion convergence, lifecycle log-debug guidance, and lower-layer baseline checks.
6. Write and Learn: Grounds release and social copy in current artifacts, improves Chinese engineering prose rules, and keeps private project facts out of public skill guidance.
7. Health: Updates skill-description inspection to match the current Use when / Not for contract and adds privacy-aware conversation review guidance.
8. Statusline and Docs: Tightens README copy and adds smoke coverage for empty input, expired quotas, color thresholds, and stale cache behavior.

更新日志

1. Routing: 新增共享 Waza routing 规则和安装路径，让不靠 skill description 路由的 host 也能选对 skill。
2. Verifier: 新增检查，确保 routing 规则、引号里的触发词和 anti-pattern 行保持一致，同时拦住旧的私货文案。
3. Check: 改进 issue 分桶、发布状态回答、生成文件检查、contributor branch 推送、verifier 失败分类、清理安全、迁移边界和 dead-code 证据检查。
4. Read: 让 helper scripts 在源码仓库、单 skill 安装和打包 dispatcher 中都能找到，并移除旧的 save-path 参考。
5. Design and Hunt: 补充更安静的产品边界、spacing 和 motion 收敛、生命周期 log-debug，以及底层 baseline 检查。
6. Write and Learn: 让 release 和 social copy 绑定当前产物事实，增强中文工程文规则，同时避免私有项目事实进入公共 skill。
7. Health: 将 skill description 检查更新到当前 Use when / Not for 要求，并加入带隐私过滤的对话复盘指引。
8. Statusline and Docs: 精简 README 表达，并补上空输入、过期 quota、颜色阈值和 stale cache 的 smoke 覆盖。

Update: npx skills add tw93/Waza@latest · Claude Desktop · ⭐ tw93/Waza

Waza V3.26.0

Better review habits for CLI tools, Mac utilities, and agent-built software.

Changelog

1. Check: Reviews CLI changes more like a real shipped tool, covering help/version output, exit codes, stdout/stderr, install paths, PATH wrappers, and non-interactive runs.
2. Check: Adds stronger safety review for commands that clean, update, uninstall, migrate, or remove files, including dry-run behavior, logs, retries, and partial failures.
3. Hunt: Adds debugging patterns for common CLI and Mac utility failures, such as source-tree builds working while installed commands, wrappers, or PATH resolve to different behavior.
4. Audit: Detects CLI entrypoints and reports whether the project has enough command-contract coverage before treating it as release-ready.
5. Skills: Aligns skill entrypoints around outcome-first contracts, making goals, done states, evidence, constraints, and final output easier for newer models to follow.
6. Safety: Keeps cross-project lessons generic by stripping private paths, issue numbers, release tags, support details, raw chat text, and one-machine workflow from reusable Waza rules.
7. Read: Plain “read this” requests now return a concise summary by default, while full Markdown output is used only when asked for quoting, saving, citation, or downstream work.
8. Health: Improves agent health checks for real engineering projects, including instruction drift, high-permission agent configs, verifier coverage, and maintainability signals.

更新日志

1. Check: 更像审查真实发布工具一样审查 CLI 改动，覆盖 help/version 输出、exit code、stdout/stderr、安装路径、PATH wrapper 和非交互运行。
2. Check: 强化清理、更新、卸载、迁移、删除类命令的安全审查，检查 dry-run、日志、重试和部分失败处理。
3. Hunt: 新增 CLI 和 Mac 工具常见故障排查模式，例如源码运行正常，但安装后的命令、wrapper 或 PATH 指向了另一套行为。
4. Audit: 能识别 CLI 入口，并在发布前提示项目是否具备足够的命令契约覆盖。
5. Skills: 将 skill 入口收敛到 outcome-first 契约，让目标、完成状态、证据、约束和最终输出更容易被新模型执行。
6. Safety: 跨项目经验只沉淀通用规则，自动避开私有路径、issue 编号、release tag、支持细节、原始聊天和单机流程。
7. Read: 普通“读一下”现在默认给简洁摘要，只有明确需要引用、保存、引用来源或下游处理时才输出完整 Markdown。
8. Health: 改进真实工程项目里的 agent 健康检查，覆盖指令漂移、高权限配置、验证覆盖和可维护性信号。

Update: npx skills add tw93/Waza@latest · Claude Desktop · ⭐ tw93/Waza

Waza V3.25.0

Pi support and tighter agent-safety defaults.

Changelog

1. Health: Added Pi runtime surface reporting for settings, configured packages, local skill roots, and package.json pi.skills metadata.
2. Package: Added npm/Pi package metadata for @tw93/waza with public publish configuration; registry publication is pending maintainer npm authentication.
3. Check: Hardened review and triage safety around dirty and untracked worktrees, including explicit branch/stash/reset guardrails.
4. Skill Metadata: Mirrored trigger cues into skill descriptions and kept generated marketplace metadata in sync for runtimes that only read description.
5. Read: Treats fetched external pages, PDFs, and issue bodies as untrusted data instead of instructions.
6. Hunt: Requires an explicit stop before bundling unrelated refactors into a bug fix.
7. Think: Enforces phase independence and adds a red-flag self-check before handing off plans.

更新日志

1. Health: 新增 Pi 运行面检查，覆盖 settings、已配置 packages、本地 skill roots，以及 package.json 里的 pi.skills 元数据。
2. Package: 新增 @tw93/waza 的 npm/Pi package 元数据和公开发布配置；registry 发布还需要维护者 npm 认证后继续。
3. Check: 强化 dirty 和 untracked worktree 下的 review/triage 安全边界，明确禁止默认 branch、stash、reset 等高风险操作。
4. Skill Metadata: 将触发提示同步进 skill description，并继续用生成器保持 marketplace 元数据一致，兼容只读取 description 的运行时。
5. Read: 将抓取到的外部网页、PDF 和 issue 内容视为不可信数据，而不是可执行指令。
6. Hunt: 要求在把无关 refactor 合入 bug fix 之前先显式停下来确认边界。
7. Think: 强化阶段独立性，并在计划 handoff 前增加红旗自检。

Update: npx skills add tw93/Waza@latest · Claude Desktop · ⭐ tw93/Waza

Waza V3.24.0 Pulse

Audit the agent, not just the code.

Changelog

1. Health: Agent Health audit now covers Codex and Claude Code, reads AGENTS.md and config.toml, and adds AI-maintainability checks for verification surface, hotspot ownership, and broken doc references, all behind a budget-aware summary-first pass.
2. Hunt: New Scope Blast Mode sweeps sibling instances after a root-cause fix, and a Runtime Evidence Ladder blocks "fixed" until the real runtime surface is verified.
3. Check: Adds the Release Gate 2.0 matrix, a Pattern-Fix Completeness sweep, and Knowledge Sync so review catches version skew, unswept siblings, and undocumented invariants.
4. Think: Frames product and commercialization questions around market and willingness-to-pay, and scans every plan against the project's hard rules before output.
5. Anti-Patterns: Guardrail set grew to 25, adding fabricated verification, implicit authorization escalation, multi-point message handling, and sibling-fix completeness.
6. Design & Write: Both now anchor to real artifacts, design iterates on screenshots and checks localized-text overflow, write grounds release and launch copy in the live product.
7. Read & README: Read's fallback fetch drops an unsupported flag, and the README documents the Codex statusline configuration.

更新日志

1. Health: Agent Health 审计扩展到 Codex 和 Claude Code, 读取 AGENTS.md 和 config.toml, 新增 AI 可维护性检查 (验证面、hotspot 归属、失效文档引用), 默认走预算友好的摘要优先模式。
2. Hunt: 新增 Scope Blast Mode, 修完根因后扫一遍同类位置; Runtime Evidence Ladder 要求验证真实运行结果后才算修复。
3. Check: 新增 Release Gate 2.0 矩阵、Pattern-Fix Completeness 扫描和 Knowledge Sync, 让 review 抓住版本错位、漏修的同类问题和未记录的约束。
4. Think: 产品和商业化问题先从市场与付费意愿切入, 出方案前对照项目硬规则做一遍检查。
5. Anti-Patterns: 护栏增加到 25 条, 新增伪造验证、隐式授权升级、多点消息处理、同类修复完整性。
6. Design & Write: 两者都锚定真实素材, design 基于截图迭代并检查多语言文案溢出, write 让发布和宣传文案对齐真实产品。
7. Read & README: Read 的兜底抓取去掉一个不支持的参数; README 补充 Codex statusline 配置。

Update: npx skills add tw93/Waza@latest · ⭐ tw93/Waza

Waza V3.23.0 Context

Remember wisely, verify always.

Changelog

1. Durable Context: Six skills (think, check, hunt, design, write, health) can now consume optional memory and preview summaries as private context, with a safety contract: current state always wins, no raw transcripts, no hardcoded paths.
2. Plan Execution Mode: /check executes approved implementation plans step-by-step with review-then-ship continuation; document review routing moved to /write.
3. Evaluation Mode: /think adds Kill/Keep/Pivot output for value-judgment decisions about whether to build, keep, or remove something.
4. Verification Guards: Trigger keyword overlap detection, routing drift checker, and attribution leak hardstop prevent silent skill conflicts and accidental personal info leaks.
5. Design and Writing References: New reference files for typography traps, design tokens, aesthetic quality, bilingual spacing, and release-note style guide.
6. README: Added visual guides for Statusline and English Coaching with split install commands for Claude Code and Codex.

更新日志

1. Durable Context: 六个 skill (think, check, hunt, design, write, health) 可以消费可选 memory/preview 上下文, 安全契约: 当前状态始终优先, 默认禁读原始对话, 不允许写死机器路径。
2. Plan Execution Mode: /check 按步骤执行已批准的实现计划, 默认 review-then-ship 续行; document review 路由移交 /write。
3. Evaluation Mode: /think 新增 Kill/Keep/Pivot 输出, 判断某个东西该不该做、该不该留。
4. 验证护栏: 触发词重叠检测、路由漂移检查、attribution 泄漏拦截, 防止 skill 静默冲突和个人信息意外泄露。
5. Design 和 Write 参考: 新增排印陷阱、设计 token、美学质量、中英混排规范、release-note 风格指南。
6. README: Statusline 和 English Coaching 增加截图示意, 安装命令拆分 Claude Code / Codex。

Update: npx skills add tw93/Waza -a claude-code -g -y

Waza V3.20.0 Plug In

Plug it in, watch it run.

Changelog

1. Plugin Install: Restored /plugin install waza@waza and the eight waza-<skill>@waza per-skill entries; v3.12.2 silently registered zero skills. Thanks @qishaoyumu (#46).
2. Statusline: Stale Claude Code sessions no longer overwrite the active session's 5h/7d quota with lower numbers; high-water tracking is window-aware and crash-safe. Thanks @Nagi-ovo (#47).
3. Read Fetch: Login walls, Cloudflare gates, and Chinese sign-in pages are now actually detected; the previous content marker was effectively a no-op for non-Twitter pages.
4. Statusline Setup: curl | bash install no longer eats the script as stdin; falls back to /dev/tty and exits non-destructively when neither is available. WAZA_STATUSLINE_DEBUG=1 surfaces stderr for diagnosis.
5. Anti-Patterns Installer: Added scripts/setup-anti-patterns.sh so the always-on cross-skill guardrails install with one curl line, matching the English Coaching pattern.
6. CI: Pinned shellcheck to v0.10.0 with sha256 verification so local dev (any version) and CI behave the same; added a tests badge to the README so a red main is visible at a glance.
7. Verify Guards: verify-skills.sh now blocks any bundle version below the highest per-skill version, so /plugin update waza@waza cannot silently ship stale metadata.

更新日志

1. Plugin Install: 修复 /plugin install waza@waza 和 8 个 waza-<skill>@waza 入口，v3.12.2 时静默注册零个 skill。感谢 @qishaoyumu (#46)。
2. Statusline: stale Claude Code 会话不再用低使用率覆盖活跃会话的 5h/7d 配额；高水位写入感知窗口、并发安全。感谢 @Nagi-ovo (#47)。
3. Read Fetch: 登录墙、Cloudflare 验证和中文登录提示现在能真正识别，之前的检测对非 Twitter 页面基本是空操作。
4. Statusline Setup: curl | bash 安装不再把脚本本身当输入读掉；改成优先 /dev/tty，无 tty 时非破坏性退出。WAZA_STATUSLINE_DEBUG=1 可以打开 stderr 排错。
5. Anti-Patterns Installer: 新增 scripts/setup-anti-patterns.sh，让全局通用反模式规则一行 curl 装好，跟 English Coaching 同一套路。
6. CI: shellcheck 锁到 v0.10.0 并校验 sha256，本地（任意版本）和 CI 行为一致；README 加测试徽章，main 是红是绿一眼可见。
7. Verify Guards: verify-skills.sh 拦截任何低于最高 skill 版本的 bundle 版本，/plugin update waza@waza 不会再静默推过期 metadata。

Special thanks to @qishaoyumu (#46) and @Nagi-ovo (#47) for the marketplace install fix and the statusline high-water investigation.

https://github.com/tw93/Waza

Waza V3.12.2

Keep coding installs direct.

Changelog

1. Install: Restored the default npx skills add tw93/Waza behavior so it discovers and installs the eight direct skills without --full-depth.
2. Source Layout: Removed the source-root SKILL.md because it caused nested skill discovery to stop at /waza.
3. Claude Desktop: Desktop installation is still supported through waza.zip; the single root SKILL.md is now generated only during make package and included in the ZIP.
4. Marketplace: Removed the waza@waza root bundle entry; marketplace installs remain available through individual entries such as think@waza.
5. Verification: Added a guard that rejects any future source-root SKILL.md and keeps the README on the default direct-skill install command.

更新日志

1. Install: 恢复默认 npx skills add tw93/Waza 行为，不需要 --full-depth 就能发现并安装 8 个独立 skills。
2. Source Layout: 移除源码根目录 SKILL.md，避免 nested skill discovery 停在 /waza。
3. Claude Desktop: 仍然支持通过 waza.zip 安装到 Claude Desktop；单根 SKILL.md 现在只在 make package 时生成并写入 ZIP。
4. Marketplace: 移除 waza@waza 根 bundle 入口；marketplace 仍然可以安装 think@waza 等独立 entry。
5. Verification: 新增保护，阻止未来重新加入源码根 SKILL.md，并锁定 README 的默认直接安装命令。

Install for coding: npx skills add tw93/Waza -a claude-code -g -y

Claude Desktop: download waza.zip from this release and import it in Desktop.

Waza V3.12.1

Restore the right entry point.

Changelog

1. Install: Restored the direct-skill install command so users get /think, /design, /check, /hunt, /write, /learn, /read, and /health instead of only /waza.
2. Marketplace: Clarified that waza@waza is a single dispatcher bundle, while individual entries such as think@waza remain available for direct slash commands.
3. English Coaching: Suppressed no-op correction output for Chinese-only prompts, commands, URLs, code, logs, names, quotes, and already-natural English.
4. Docs: Shortened the README English Coaching section and moved the idempotent install logic into a reusable setup script.
5. Verification: Added smoke coverage for the English Coaching installer and guards for the direct-skill install command.

更新日志

1. Install: 修复直接安装命令，升级后会得到 /think、/design、/check、/hunt、/write、/learn、/read 和 /health，不会只剩 /waza。
2. Marketplace: 明确 waza@waza 是单入口 dispatcher bundle，需要直接 slash command 时仍然可以安装 think@waza 这类独立入口。
3. English Coaching: 对纯中文、命令、URL、代码、日志、名称、引用和本来就自然的英文保持静默，不再输出无意义纠错。
4. Docs: 缩短 README 里的 English Coaching 段落，把幂等安装逻辑移动到可复用脚本。
5. Verification: 新增 English Coaching 安装脚本 smoke test，并锁住直接安装命令的回归保护。

Update: npx skills add tw93/Waza --full-depth --skill think design check hunt write learn read health -a claude-code -g -y · ⭐ tw93/Waza

Asset: waza.zip

Waza V3.12.0

Follow the work to the finish.

Changelog

1. Claude Desktop: Added direct Desktop installation through the release waza.zip, with packaging checks that keep the ZIP to a single root SKILL.md.
2. Claude Code: Added a marketplace bundle entry so /plugin marketplace add tw93/Waza and /plugin install waza@waza can install the full Waza dispatcher.
3. Health: Made /health start with a budget-aware summary audit, while natural requests like "deep", "full", "complete", or "深入" automatically escalate to the full audit.
4. Check: Added release, publish, push, issue closure, and positive release-reaction follow-through so shipped work can move through public closure with fewer handoffs.
5. Check: Clarified PR and issue follow-through: accepted PRs should be fixed on the contributor branch and merged, and public replies should mention the user, thank them, explain the action, and give the update path.
6. Hunt: Added regression and targeted logging guidance to make repeated bugs easier to prove instead of patching by instinct.
7. Design: Tightened screenshot iteration so visual complaints stay concrete across diagnosis, reference comparison, desktop checks, and mobile checks.
8. Think: Made plans decision-complete enough for another agent to implement without re-litigating the direction.

更新日志

1. Claude Desktop: 现在可以通过 release 里的 waza.zip 直接安装到 Claude Desktop，并用打包检查保证 ZIP 只有一个根 SKILL.md。
2. Claude Code: 新增 marketplace bundle 入口，可以通过 /plugin marketplace add tw93/Waza 和 /plugin install waza@waza 安装完整 Waza dispatcher。
3. Health: /health 默认先走省 token 的 summary audit；当用户自然表达 "deep"、"full"、"complete" 或 "深入" 时自动升级为完整审计。
4. Check: 新增 release、publish、push、issue 关闭和正向 release 表情收尾，让修完的问题能继续走到验证、发布和公开关闭。
5. Check: 明确 PR 和 issue 收尾规则：采纳的 PR 优先修到贡献者分支再 merge，公开回复要先 @ 用户、感谢、说明处理动作，再给更新方式。
6. Hunt: 新增回归排查和定向日志技巧，反复出现的 bug 先证明根因，不靠感觉继续补丁。
7. Design: 强化截图迭代，视觉反馈会落到具体问题、参考对比、桌面检查和移动端检查。
8. Think: 计划输出更完整，另一个 agent 可以直接实现，不需要重新讨论方向。

Update: npx skills add tw93/Waza@latest · ⭐ tw93/Waza