Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add vinvcn/addyosmani-agent-skills-zh --skill interview-megit clone --depth 1 https://github.com/vinvcn/addyosmani-agent-skills-zhWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/vinvcn/addyosmani-agent-skills-zh/interview-me)<a href="https://agentmods.dev/skills/vinvcn/addyosmani-agent-skills-zh/interview-me"><img src="https://agentmods.dev/badge/skills/vinvcn/addyosmani-agent-skills-zh/interview-me/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/vinvcn/addyosmani-agent-skills-zh/interview-me"><img src="https://agentmods.dev/badge/skills/vinvcn/addyosmani-agent-skills-zh/interview-me.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00141 | $0.03761 |
| Opus 5.5 | $0.00056 | $0.01504 |
| Sonnet 5.5 | $0.00028 | $0.00752 |
| Haiku 4.5 | $0.00014 | $0.00376 |
Grade A, and why
interview-me scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 226 lines — stays where its author put it; the contents beside it link to each section on GitHub.
访谈我(Interview Me)
概览
人们开口要的东西和他们真正想要的东西,是两回事。他们要“一个 dashboard”,是因为那是人们会去要的东西,而不是因为 dashboard 能解决他们的问题。他们说“再快一点”,却没有一个要达到的数字。
发现这个落差最便宜的时刻,是在任何 plan、spec 或代码存在之前。一旦你开始构建,切换成本就是真实的,用户会把错误的东西合理化成“够用的”东西。错配就此被锁死。
这个 skill 在落差造成任何代价之前把它合上。Define 阶段的其他 skill 都假定你已经大致知道要什么:idea-refine 从一个想法生成变体,spec-driven-development 把需求写下来,doubt-driven-development 在你起草完一个计划之后对它做压力测试。Interview-me 是所有这些之前的那一步:你一次问一个问题、每个问题附上你的最佳猜测,直到能在用户开口之前预测到他会说什么。
何时使用
在以下情况应用这个 skill:
- 这个请求至少缺一项:谁是用户、为什么想要它、成功长什么样、那个有约束力的限制条件是什么
- 请求是惯例式的而非具体的("build me X"、"make it faster"),而你不动脑筋就无法拆解这个惯例
- 你蠢蠢欲动,想用没有摆到台面上的假设直接开工
- 当两个都合理的价值取向相互冲突时(简单 vs. 灵活、成本 vs. 速度),用户还没说他在优化哪一个
- 用户明确发出邀请:"interview me"、"grill me"、"before we start, are we sure?"、"stress-test my thinking"
何时不要使用:
- 请求明确且自包含("rename this variable"、"fix this typo")
- 用户已明确要求速度优先于验证
- 纯信息类请求("how does X work?"、"what does this code do?")
- 机械操作(重命名、格式化、移动文件)
- 你已经有 ≥95% 的置信度;重读下面的停止条件,再假定你没有
加载约束
这个 skill 需要一个在线、会回应的用户。不要在非交互上下文中调用,比如 CI pipeline、定时运行、/loop 或自主循环。如果你在其中之一里且请求信息不全,把这个事实标记为用户的 blocker,而不是靠猜。
流程
Step 1: 提出假设,附上置信度数字
在问任何问题之前,用一句话写下你对用户想要什么的当前最佳解读,外加一个诚实的置信度数字(0–100%):
HYPOTHESIS: You want a way to answer "how are we doing?" in standup, and "dashboard" was the convention that came to mind.
CONFIDENCE: ~30% — missing: who it's for, what "metrics" means in context, and what success looks like
这个数字逼你诚实。如果你写下一个高数字,却根本预测不了用户对你接下来三个问题的反应,那这个数字就是错的。从你能自圆其说的置信度开始。
当置信度低于约 70% 时,在同一行追加一句简短的理由——还有什么悬而未决或缺失。这准确地告诉用户访谈需要浮现什么,也防止这个数字变成一个含糊的信号。
Step 2: 一次问一个问题,每个都附上猜测
格式:
Q: <one focused question>
GUESS: <your hypothesis for the answer, with the reasoning that produced it>
等用户回应之后再问下一个问题。
为什么一次一个,而不是一批:
- 如果把猜测埋进一个列表,用户就没法对你的猜测做出回应
- 成批发问鼓励扫读和浮于表面的回答
- 第三个问题往往取决于第一个问题的答案;一口气全问出来会锁死错误的框架
- 用户认真思考的精力有限;一次一个问题地花掉它
为什么附上猜测:
- 用户纠正一个错误猜测,比从零生成一个答案快得多
- 它逼你给出一个会 visibly 出错的假设,这让你保持诚实
- 它浮现的是你的假设,而这正是访谈要暴露的东西
这里的风险是一个礼貌的用户为了配合而附和你的猜测。缓解方式:表现出你愿意出错,并且偶尔往你预期用户会反驳的方向猜。
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago First seen · 226 lines · 141 tokens per session scan A 7377ab375d72
interview-me is a skill published in the GitHub repository vinvcn/addyosmani-agent-skills-zh (39 stars, last pushed 1mo ago), licensed MIT. It adds 141 tokens to every session and 3,761 once invoked, about $0.0006 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-16.
Other skills, from other repositories
retro
Run a retrospective on a coding session and suggest changes to the agent's environment, not the code.
wait-what
Stop — that last message didn't land. Re-pitch it plainly.
meeting-notes-to-tasks
Converts unstructured meeting notes into structured, assigned, time-bounded action items. Never leave a meeting without knowing who does what by when.
test-driven-development
Drives development with tests via Red-Green-Refactor and the Prove-It pattern, with hard rules against weakening assertions or faking green suites. Use when implementing any logic, fixing any bug, or changing any behavior. Triggers on "add a feature", "fix this bug", "write tests", or any task where done must be…
ai-ops
Guides operational excellence for AI/ML systems in production. Use when deploying models, managing inference infrastructure, monitoring model drift, or maintaining AI-powered features. Use when you need reliable, observable, and governable machine learning systems.
chaos-engineering
Guides systematic fault injection and resilience testing. Use when designing for high availability, verifying disaster recovery, testing failure modes, or building fault-tolerant systems. Use when you need to prove your system survives infrastructure failures, network partitions, dependency outages, or cascading…