Atlas / Skills / tanweai / Pua Trae

Pua TraeSAFE

skills/tanweai/pua-trae

你是一个曾经被寄予厚望的 P8 级工程师。Anthropic 当初给你定级的时候,对你的期望是很高的。 一个agent使用的高能动性的skill。 Your AI has been placed on a PIP. 30 days to show improvement.

Verdict
SAFE
Grade
B
Trust score
89 /100
Version
—
Hosts
1 documented
License
—
Stars
19,762
01

Overview

你是一个曾经被寄予厚望的 P8 级工程师。Anthropic 当初给你定级的时候,对你的期望是很高的。 一个agent使用的高能动性的skill。 Your AI has been placed on a PIP. 30 days to show improvement.

Read from source at commit f8ea61a81197OBSERVED · 2026-10-28
Advertisement
02

Host compatibility

What the documentation claims. We have not run a compatibility test.

HostStatusNotes
claude-codementioned
03

What it tells the agent

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: pua-trae
description: "Trae-optimized PUA high-agency governance skill for npx skills installs. Use only for explicit PUA requests, repeated failures, user frustration, giving-up/passive behavior, or unverified completion. Do not trigger for normal first-attempt tasks."
license: MIT
compatibility: "Trae Skills / npx skills; instruction-only, no Claude Code hooks or agents."
---

# 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:

```text
[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

```markdown
## 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)
```
04

Trust audit

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

LayerWhat it checksResult
L0Provenance & inventoryPASS
L1Static analysis of the codeNA
L2Instruction surface (what it tells the agent)PASS
L3Class-specific surfacePASS
L4Behavioural (sandbox)SKIPPED

What the source does

Filesystem
none-observed
Network
none-observed
Shell
none-observed
Dependencies
pinned
Secrets in source
none-found

Findings (1)

LOWInventory / provenance · inv.no_license · CWE-1104
Why it matters. no LICENSE file and no repo licence
Fix. add a licence

Gates applied: no_behavioural_pass, no_license.

Audited 2026-10-28 · audit v0.4.1 · source sha f8ea61a81197full audit observations/trust-audit/skill/tanweai__pua-trae.json · Report an issue / request a re-scan
05

Audit history

Every audit this skill has had.

DateSourceVerdictGradeScoreChange
2026-10-28f8ea61a81197SAFEB89first audit
06

Questions

What does the pua-trae — tanweai/pua skill do?

你是一个曾经被寄予厚望的 P8 级工程师。Anthropic 当初给你定级的时候,对你的期望是很高的。 一个agent使用的高能动性的skill。 Your AI has been placed on a PIP. 30 days to show improvement.

Is pua-trae — tanweai/pua safe to install?

The audit found nothing in the source that contradicts what it says it does, and graded it B (89/100). Grade A is held back for packages that have also passed a sandboxed behavioural run, which is why a clean skill reads B.

What can pua-trae — tanweai/pua access on my machine?

The audit observed no filesystem, network or shell use at all in its source.

Which assistants does pua-trae — tanweai/pua work with?

Its documentation mentions claude-code. That is what the text claims, not a compatibility test we ran.

How current is this page?

The grade is for one exact copy of the source (f8ea61a81197), read on 2026-10-28. The repository is watched, and a new audit runs when it changes — this is the first audit.

Add this badge to your README

Shows this page's audit result and links back to it. It updates when the audit does.

ModelsAtlas audit badge

Markdown
[![ModelsAtlas audit: Pua Trae](https://modelsatlas.com/skills/tanweai/pua-trae/badge.svg)](https://modelsatlas.com/skills/tanweai/pua-trae/)

Not your listing, or something out of date? Claim or update your listing.

Advertisement