AI代理的简洁输出技能。简短的回复,不剥离技术细节。
概览
AI代理的简洁输出技能。简短的回复,不剥离技术细节。 “当然!我很乐意帮你。你遇到的问题很可能是认证中间件出了问题,令牌过期检查用错了比较操作符......” 认证中间件出了漏洞。令牌过期检查使用 < 而不是 <=。修复方法: Crisp 受到了现有简洁输出和代理行为研究的启发,包括: - Julius Brussee 的穴居人 - Matt Pocock 的穴居人技能 - 编码代理工作流程中使用的常见“无冗余”提示模式 简洁是我对这个想法的专业语气变体:正常简洁的回复,不出现表情包式输出,并仔细保留代码、错误、命令、数字和风险操作上下文。 Crisp 注重简洁且听起来自然且专业的回复。 目标不是不惜一切代价实现最大压缩。目标是在保持清晰度的同时去除填充部分,尤其是在编码和技术流程中。
概述(AI 翻译)
AI代理的简洁输出技能。简短的回复,不剥离技术细节。
之前 / 之后
没有脆皮:
“当然!我很乐意帮你解决这个问题。你遇到的问题很可能是你的认证中间件出了问题,令牌检查用错了比较操作符…”
带crisp:
认证中间件的 bug。令牌到期检查使用
技术细节相同。词语更少。
署名
Crisp 受到了现有简洁输出和代理行为研究的启发,包括:
- 朱利叶斯·布鲁塞的《穴居人》
- Matt Pocock的穴居人技能
- 编码代理工作流程中使用的常见“无冗余”提示模式
简洁是我对这个想法的专业语气变体:正常简洁的回复,不出现表情包式输出,并仔细保留代码、错误、命令、数字和风险操作上下文。
为什么清脆
Crisp 注重简洁且听起来自然且专业的回复。
目标不是不惜一切代价实现最大压缩。目标是在保持清晰度的同时去除填充部分,尤其是在编码和技术流程中。
本页由 AI 翻译,技术细节请以原文为准。
README
以下为原文 README 节选,完整内容请查看原文。
crisp
A terse-output skill for AI agents. Shorter replies without stripping technical details.
Before / After
Without crisp:
“Sure! I’d be happy to help you with that. The issue you’re experiencing is likely caused by a problem in your authentication middleware where the token expiry check is using the wrong comparison operator…”
With crisp:
Bug in auth middleware. Token expiry check uses
<not<=. Fix:if (now <= expiry) { ... }
Same technical details. Fewer words.
Attribution
crisp was inspired by existing terse-output and agent-behavior work, including:
- Caveman by Julius Brussee
- Caveman skill by Matt Pocock
- Common “no fluff” prompting patterns used in coding-agent workflows
crisp is my professional-tone variation of this idea: normal concise replies, no meme-style output, and careful preservation of code, errors, commands, numbers, and risky-operation context.
Why crisp
crisp focuses on concise replies that still sound normal and professional.
The goal is not maximum compression at all costs. The goal is to remove filler while preserving clarity, especially in coding and technical workflows.
crisp is designed to stay terse by default, but use clearer wording when compression could make the answer unsafe or ambiguous.
For destructive operations, security warnings, and irreversible actions, crisp instructs the agent to use clear full-sentence warnings before returning to concise output. The goal is faster reading without compressing the parts where clarity matters.
// Destructive op — crisp uses clear prose first:
Warning: This permanently destroys all data in `users`. Cannot be undone.
DROP TABLE users;
Verify backup first. ← concise mode resumes here
Install
crisp follows the Agent Skills specification. It works with any agent that supports skills.
npx skills add shubhamv123/crisp
Any other agent / browser:
Paste the contents of SKILL.md at the start of your conversation:
“Follow these instructions for this entire conversation: [paste SKILL.md]”
Usage
| What you type | What happens |
|---|---|
crisp mode |
Activates crisp |
go crisp |
Activates crisp |
/crisp |
Activates crisp |
be brief |
Activates crisp |
cut the fluff |
Activates crisp |
stop crisp |
Back to normal |
normal mode |
Back to normal |
explain in detail |
Back to normal |
Once active, crisp stays on for the entire conversation — no need to re-trigger every message.
What crisp reduces
- Unnecessary openings: “sure”, “happy to help”, “great question”
- Soft filler: “basically”, “actually”, “simply”, “just”
推荐工具
换一个关键词,或者移除筛选条件。
安装
npx skillfish add shubhamv123/crisp