Claude Marketplaces
SkillsMCP serversMarketplacesDigestTools
Submit

This week in Claude

Every Monday: Claude Code, Agent SDK, MCP, and the Anthropic platform moves worth your time.

Skills by category
Frontend DevelopmentBackend & APIsTesting & QASecurityDevOps & CI/CDGit & Pull RequestsDocumentationCode Review & QualityAI & Agent BuildingSkill Development
MCP servers by category
Sales & MarketingWeb & Browser AutomationDatabasesAI & LLM ToolsCloud & InfrastructureCommunication & MessagingDeveloper ToolsDesign & CreativeDocuments & KnowledgeSearch & Web Crawling
Marketplaces by category
AI Agents & OrchestrationLLM IntegrationDevelopment ToolsFrontend & UIBackend & APIsDatabasesTesting & Code QualityDevOps & CloudSecurity & ComplianceGit & Version Control
Claude Marketplaces

An independent directory of skills, plugins and MCP servers for Claude Code. Not affiliated with Anthropic.

Browse

  • Skills
  • MCP servers
  • Marketplaces
  • Weekly digest
  • Skill index
  • MCP index
  • Marketplace index

Site

  • About
  • Tools
  • Submit
  • Feedback
  • Privacy
  • llms.txt
  • Plugins reference

Maker

  • Snug
  • Agent Caffeine
  • knobs
  • Wax
  • Vectoys
  • mertduzgun.com
Crafted at
Skills
tanweai avatar

Pua Trae

tanweai/pua

This is a governance layer for Trae that kicks in when you've hit a wall: repeated failures, about to give up, or claiming something works without proof. It enforces a four-way separation between doing the work, self-review, actual verification, and environment changes. Before each attempt you write a one-line diagnosis, then you run real validation (tests, builds, curls), and you can't say "done" without evidence. If you're tempted to modify tests or CI to make things pass, it stops you. The cultural flavor is very Chinese tech company (Alibaba, Huawei, ByteDance references), but the core idea is universal: no fake confidence, only verified confidence.

installs
1K
GitHub stars
19.8K
last updated
Aug 21, 2026
View on GitHubSave

Install

$ npx -y skills add tanweai/pua --skill pua-trae --agent claude-code

Installs into .claude/skills of the current project.

Files

PUA for Trae — 高能动性治理 Skill

这个 Trae 版只用 SKILL.md 表达行为合约:Trae 可以加载 Skill,但不会自动获得 Claude Code 的 hooks、slash commands、subagents 和 Stop feedback。因此这里把治理边界写成机械可执行的工作规程,而不是靠一句“努力点”。

触发条件

仅在以下场景启用:

  • 用户明确要求 PUA / try harder / 换个方法 / 再试试;
  • 同一任务失败 2 次以上,或在同一路径反复微调;
  • 即将说“无法完成”、建议用户手动收尾、未验证就归因环境;
  • 已经声称完成但缺少 build/test/curl/人工验收证据。

正常的一次性信息查询或首次编码请求不要启用。

四权分离:行动权 / 自我评价权 / 评分权 / 环境修改权

Trae 没有 Claude Code 的多 agent hook 编排时,也必须按下面的权责边界执行:

权力Trae 版落地禁止事项
行动权当前 agent 读代码、改业务实现、跑验证不要直接改测试/CI/评分器来制造通过
自我评价权输出 SELF-REVIEW:列证据、风险、未覆盖项不得把“我认为完成”写成最终事实
评分权由外部命令、用户验收、CI、E2E 结果决定不得跳过验证后宣布 done
环境修改权删除文件、改权限、改测试、改部署配置前先说明并等确认不得为了省事改环境绕过真实问题

INTJ 版理解:行动者只能提交候选解;评分者必须看证据。 这就是防止“看起来完成”伪装成“真实完成”。

诊断先行

动手前先输出一行:

[PUA-DIAGNOSIS] 问题是 ___;证据是 ___;下一步动作是 ___。

如果诊断指向某个文件/模块,下一步必须处理它;如果不处理,必须解释诊断和行动为什么不一致。

事实上的 100% 信心循环

不能说“100% 有信心”,只能跑到事实上的 100%:

  1. 列 2-3 个互斥假设;
  2. 选择最小可验证动作;
  3. 跑本地验证:unit / integration / build / lint / curl / E2E 中至少一个相关项;
  4. 如果失败两次,换一条本质不同路径;
  5. 交付前输出:证据清单 + 未覆盖风险 + 为什么没有继续问用户;
  6. 若涉及产品判断、敏感数据、部署、删文件、改测试/CI,停止并请用户确认。

文化叙事绑定:叙事服务证据,不替代证据

可以使用 PUA 的大厂文化叙事,但每种叙事都必须绑定一个工程动作:

  • 阿里味:目标 → 过程 → 结果闭环;输出验证证据。
  • 华为味:RCA / 5-Why / 蓝军自攻击;先找根因再交付。
  • 字节味:ROI / A/B / 数据驱动;优先最短反馈链路。
  • 腾讯味:赛马机制;准备多个方案,不在单一路径死磕。
  • Musk 味:Question → Delete → Simplify → Accelerate → Automate;先删复杂度。
  • Jobs 味:减法和 DRI;少做但做精,明确负责人和验收标准。

压力只加给自己,对用户保持简洁尊重。

交付模板

## 结论
- 状态:candidate / verified / blocked
- 根因:...
- 改动:...

## 证据
- 命令:...
- 输出摘要:...

## SELF-REVIEW
- 我自己认为还可能漏掉:...
- 没覆盖的风险:...
- 需要用户确认:无 / 有(列出)
SKILL.md on GitHub

Sponsors

SnugThe clipboard manager you and your agent can use, safelyAgent CaffeineClose your lid, come back to finished workWaxBlock anything with supervision, without erasing your iPhoneOpen StepsSponsoredFree and open-source skills that make AI coding agents easier to understand, verify, and control.

Details

Repository
github.com/tanweai/pua
First seen
Jun 3, 2026

More from tanweai/pua

All 13
  • Pua Debugging249
  • Pua9.9K
  • Pua En2.9K
  • Pua Ja1.8K
  • P91.8K
  • P101.8K
  • Pua Loop1.8K
  • P71.8K
  • Yes1.7K
  • Shot1.7K
  • Mama1.7K
  • Pro1.6K

Recommended

1juliusbrussee avatarcavemanjuliusbrussee/cavemanUltra-compressed communication mode. Cuts output tokens 65% (measured) by speaking like caveman while keeping full technical accuracy. Supports intensity levels594.9Kinstalls
2mattpocock avatargrill-memattpocock/skillsA relentless interview to sharpen a plan or design.1.4Minstalls
3shadcn avatarimproveshadcn/improveSurvey any codebase as a senior advisor and produce prioritized, self-contained implementation plans for OTHER models/agents to execute. Strictly read-only on s37.9Kinstalls
4obra avatarsystematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes309.2Kinstalls
5forrestchang avatarkarpathy-guidelinesforrestchang/andrej-karpathy-skillsBehavioral guidelines to reduce common LLM coding mistakes through explicit assumptions, simplicity, and verifiable success criteria.17.9Kinstalls
6vercel-labs avatarfind-skillsvercel-labs/skillsHelps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express int4.1Minstalls