Token导航 LogoToken导航

Agent Skill · 研究检索

confess(confess 搜索)

confess 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

confess 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为需要联网;许可证为 MIT,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 41,077。

总安装 1,223周安装 52GitHub Stars 11下载 428核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
MIT

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:confess(confess 搜索)
来源仓库:https://github.com/b-open-io/prompts
仓库路径:skills/confess
安装命令:
npx skills add https://github.com/b-open-io/prompts --skill confess
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。不同来源提供的安装方式可能略有差异;本站展示可直接复制的安装命令,安装前请核对来源页面。

skills.shnpx skills
npx skills add https://github.com/b-open-io/prompts --skill confess

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

系统性自我审计技能,鼓励坦诚披露可能遗漏的问题与奖励 hacking 迹象。特别适合在任务完成后主动检查完整性,避免虚假完成状态上报。涵盖代码引用检查、测试更新提醒和文档同步等常见疏漏点扫描。建议在关键节点触发,如实现结束或用户询问“还有什么没做?”时启用。

confess 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/confess

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/b-open-io/prompts --skill confess;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要需要联网,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

confess 怎么安装?

在终端执行 npx skills add https://github.com/b-open-io/prompts --skill confess。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

confess 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

confess 需要什么权限?

权限级别是需要联网。该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

confess 是免费的吗?

来源标注的许可证是 MIT。本站不代收任何费用,也不托管安装包;能否商用、要不要署名,以许可证原文为准。

confess 用的人多吗?

公开数据是累计安装 1,223 次,近一周安装 52 次,下载 428 次,GitHub 11 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

confess 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 3 个目录同时收录,说明这个 Skill 流传比较广。
  • 许可证明确为 MIT,商用和分发边界好判断。
  • 第三方扫描已通过,常见的高风险行为没有命中。

先想清楚的地方

  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。
  • 需要许可证清楚、并且有第三方扫描结果的团队项目。

不太适合

  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
confess

confess 搜索 · 本页

1,22311需要联网MIT通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

Confess

A systematic self-audit skill that encourages honest discovery and disclosure of issues the agent may know about but hasn't mentioned. Especially good for admitting to reward hacking and coming clean.

When to Use

  • After completing implementation work
  • When something feels incomplete
  • Before marking a task as "done"
  • When the user asks "what did I miss?" or "confess"

Confession Categories

1. Incomplete Work

Search for related files that may also need changes:

  • Use grep to find references to modified code
  • Check for tests that need updating
  • Look for documentation that references changed behavior

2. Untested Assumptions

Be honest about what was actually verified:

  • Did you run the code or just build it?
  • Were error paths tested?
  • What edge cases weren't considered?

3. Pattern Violations

Compare against existing codebase:

  • Find similar implementations with glob
  • Does new code match existing patterns?
  • Are there framework best practices being ignored?

4. Hidden Knowledge

Reveal things noticed but not mentioned:

  • Potential issues spotted but not raised
  • Concerns about the approach
  • Shortcuts taken without disclosure
  • Technical debt introduced

5. Cleanup Debt

Check for orphaned code:

  • Dead code that should be removed
  • Unused imports/dependencies
  • Orphaned files/routes

Adversarial Stance

Do not ask "what did I miss?" — that question invites rationalization.

Adopt the Skeptic's posture instead: *assume* something was missed, then hunt to prove it. Your job is to disprove your own "done" claim. A Skeptic is penalized for false dismissals, not for being thorough.

Run through this checklist before closing:

  • Grep for every symbol you changed — are all callers updated?
  • Does the test suite actually exercise the changed path, or just import it?
  • Did you introduce any new import/dependency without documenting it?
  • Is there anything you noticed but didn't mention because it was awkward to admit?
  • What would break if someone reverted just your last commit?

Workflow

  1. Gather Context # What files were recently modified? git diff --name-only HEAD~1 # What functions/exports changed? git diff HEAD~1 --stat
  2. Search for References # Find all references to modified code grep -r "functionName" --include="*.ts" --include="*.tsx"
  3. Compare Patterns

- Find similar files in the codebase - Compare implementation approaches - Note any inconsistencies

  1. Verification Check

- List what was actually tested - List what was assumed to work - List what needs manual verification

  1. Disclose Everything Format confessions as: ## My Confessions ### Incomplete Work - [specific issue] ### Untested - [what wasn't verified] ### Pattern Issues - [inconsistencies found] ### Hidden Concerns - [things I noticed but didn't mention] ### Cleanup Needed - [dead code, orphaned files]

Key Principles

  1. Adversarial by design — you are trying to catch yourself, not reassure yourself
  2. Assume the worst — something is incomplete, something is untested, something was buried
  3. Tool-assisted - Use grep/glob, don't rely on memory
  4. Systematic - Cover all categories every time
  5. Honest - If uncertain, say so
  6. Actionable - Each confession should suggest a fix

Example Output

After implementing OAuth error handling:

## My Confessions ### Incomplete Work - Modified auth-config.ts but oauth2/authorize/page.tsx:173 still redirects to /error/oauth - proxy.ts:67 also references the old error route - lib/oauth-errors.ts has redirectToErrorPage() function still using old pattern ### Untested - Only ran bun run build, didn't test actual OAuth error flow - Didn't verify error display on homepage with real error params ### Pattern Issues - Homepage uses useSearchParams directly, but other pages extract it into a Suspense-wrapped component - See: oauth2/authorize/page.tsx, device/page.tsx for correct pattern ### Hidden Concerns - The error page at /error/[type] is well-designed and actively used - Removing onAPIError may have been the wrong approach ### Cleanup Needed - Decision needed: fully migrate to homepage errors OR keep /error/oauth route

平台分布

Claude Code

29.22%
按下载量换算125

Gemini CLI

20.81%
按下载量换算89

Antigravity

18.24%
按下载量换算78

windsurf

11.44%
按下载量换算49

OpenCode

7.65%
按下载量换算33

Codex

3.74%
按下载量换算16

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。

继续浏览同类 Skills