重要前提
安装AI Skills的关键前提是:必须科学上网,且开启TUN模式,这一点至关重要,直接决定安装能否顺利完成,在此郑重提醒三遍:科学上网,科学上网,科学上网。查看完整安装教程 →
verification-before-completion by oimiragieo/agent-studio
npx skills add https://github.com/oimiragieo/agent-studio --skill verification-before-completion未经验证就声称工作已完成是不诚实的行为,而非高效。
核心原则: 先有证据,再作声明,始终如此。
违反此规则的字面意思即是违反其精神。
NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE
如果你没有在当前消息中运行验证命令,就不能声称它通过了。
BEFORE claiming any status or expressing satisfaction:
1. IDENTIFY: What command proves this claim?
2. RUN: Execute the FULL command (fresh, complete)
3. READ: Full output, check exit code, count failures
4. VERIFY: Does output confirm the claim?
- If NO: State actual status with evidence
- If YES: State claim WITH evidence
5. ONLY THEN: Make the claim
Skip any step = lying, not verifying
| 声明 | 要求 | 不足够 |
|---|---|---|
| 测试通过 | 测试命令输出:0 失败 |
广告位招租
在这里展示您的产品或服务
触达数万 AI 开发者,精准高效
| 之前的运行结果,"应该通过" |
| Linter 干净 | pnpm lint:fix 输出:0 错误 | 部分检查,推测 |
| 格式化干净 | pnpm format 输出:无更改 | 目视检查,假设 |
| 构建成功 | 构建命令:退出码 0 | Linter 通过,日志看起来没问题 |
| Bug 已修复 | 测试原始症状:通过 | 代码已更改,假设已修复 |
| 回归测试有效 | 红绿循环已验证 | 测试通过一次 |
| Agent 已完成 | VCS 差异显示更改 | Agent 报告"成功" |
| 需求已满足 | 逐行核对清单 | 测试通过 |
| 代码质量门禁 | pnpm lint:fix + pnpm format 通过 | 测试通过 |
| 借口 | 现实 |
|---|---|
| "现在应该可以了" | 运行验证 |
| "我有信心" | 信心不等于证据 |
| "就这一次" | 没有例外 |
| "Linter 通过了" | Linter 不等于编译器 |
| "Agent 说成功了" | 独立验证 |
| "我累了" | 疲惫不等于借口 |
| "部分检查就够了" | 部分证明不了什么 |
| "措辞不同所以规则不适用" | 精神重于字面 |
测试:
CORRECT: [Run test command] [See: 34/34 pass] "All tests pass"
WRONG: "Should pass now" / "Looks correct"
回归测试(TDD 红绿循环):
CORRECT: Write -> Run (pass) -> Revert fix -> Run (MUST FAIL) -> Restore -> Run (pass)
WRONG: "I've written a regression test" (without red-green verification)
构建:
CORRECT: [Run build] [See: exit 0] "Build passes"
WRONG: "Linter passed" (linter doesn't check compilation)
Lint 和格式化(阻塞门禁):
CORRECT: [Run pnpm lint:fix] [See: 0 errors] [Run pnpm format] [See: no changes] "Lint and format clean"
WRONG: "Code looks formatted" / "No obvious lint issues" / "Should be clean"
需求:
CORRECT: Re-read plan -> Create checklist -> Verify each -> Report gaps or completion
WRONG: "Tests pass, phase complete"
Agent 委托:
CORRECT: Agent reports success -> Check VCS diff -> Verify changes -> Report actual state
WRONG: Trust agent report
来自失败分析:
总是在以下情况之前:
规则适用于:
验证没有捷径。
运行命令。阅读输出。然后声明结果。
这是不可协商的。
| 反模式 | 为何失败 | 正确方法 |
|---|---|---|
| 在运行命令前声称成功 | 没有证据表明声明为真 | 运行命令,阅读完整输出,然后声明 |
| 信任之前运行的结果 | 状态可能自上次执行后已改变 | 总是在当前会话中运行新的验证 |
| 部分验证(测试了但没 lint) | Lint 失败即使测试通过也会导致 CI 失败 | 运行所有门禁:测试、pnpm lint:fix、pnpm format |
| 使用"应该"或"可能"等语言 | 暗示假设,而非验证 | 消除模糊措辞;验证然后陈述事实 |
| 跳过红绿重构循环 | 总是通过的回归测试无法捕获任何问题 | 在标记完成前验证还原后测试失败 |
开始前: 阅读 .claude/context/memory/learnings.md
完成后:
.claude/context/memory/learnings.md.claude/context/memory/issues.md.claude/context/memory/decisions.md假设中断:如果不在记忆中,就等于没发生。
每周安装数
57
仓库
GitHub Stars
20
首次出现
Jan 27, 2026
安全审计
安装于
github-copilot56
gemini-cli55
kimi-cli54
amp54
codex54
opencode54
Claiming work is complete without verification is dishonesty, not efficiency.
Core principle: Evidence before claims, always.
Violating the letter of this rule is violating the spirit of this rule.
NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE
If you haven't run the verification command in this message, you cannot claim it passes.
BEFORE claiming any status or expressing satisfaction:
1. IDENTIFY: What command proves this claim?
2. RUN: Execute the FULL command (fresh, complete)
3. READ: Full output, check exit code, count failures
4. VERIFY: Does output confirm the claim?
- If NO: State actual status with evidence
- If YES: State claim WITH evidence
5. ONLY THEN: Make the claim
Skip any step = lying, not verifying
| Claim | Requires | Not Sufficient |
|---|---|---|
| Tests pass | Test command output: 0 failures | Previous run, "should pass" |
| Linter clean | pnpm lint:fix output: 0 errors | Partial check, extrapolation |
| Format clean | pnpm format output: no changes | Visual inspection, assumption |
| Build succeeds | Build command: exit 0 | Linter passing, logs look good |
| Bug fixed | Test original symptom: passes | Code changed, assumed fixed |
| Regression test works | Red-green cycle verified | Test passes once |
| Agent completed | VCS diff shows changes | Agent reports "success" |
| Requirements met | Line-by-line checklist | Tests passing |
| Code quality gates | pnpm lint:fix + pnpm format passed | Tests passing |
| Excuse | Reality |
|---|---|
| "Should work now" | RUN the verification |
| "I'm confident" | Confidence does not equal evidence |
| "Just this once" | No exceptions |
| "Linter passed" | Linter does not equal compiler |
| "Agent said success" | Verify independently |
| "I'm tired" | Exhaustion does not equal excuse |
| "Partial check is enough" | Partial proves nothing |
| "Different words so rule doesn't apply" | Spirit over letter |
Tests:
CORRECT: [Run test command] [See: 34/34 pass] "All tests pass"
WRONG: "Should pass now" / "Looks correct"
Regression tests (TDD Red-Green):
CORRECT: Write -> Run (pass) -> Revert fix -> Run (MUST FAIL) -> Restore -> Run (pass)
WRONG: "I've written a regression test" (without red-green verification)
Build:
CORRECT: [Run build] [See: exit 0] "Build passes"
WRONG: "Linter passed" (linter doesn't check compilation)
Lint and Format (BLOCKING GATE):
CORRECT: [Run pnpm lint:fix] [See: 0 errors] [Run pnpm format] [See: no changes] "Lint and format clean"
WRONG: "Code looks formatted" / "No obvious lint issues" / "Should be clean"
Requirements:
CORRECT: Re-read plan -> Create checklist -> Verify each -> Report gaps or completion
WRONG: "Tests pass, phase complete"
Agent delegation:
CORRECT: Agent reports success -> Check VCS diff -> Verify changes -> Report actual state
WRONG: Trust agent report
From failure analysis:
ALWAYS before:
Rule applies to:
No shortcuts for verification.
Run the command. Read the output. THEN claim the result.
This is non-negotiable.
| Anti-Pattern | Why It Fails | Correct Approach |
|---|---|---|
| Claiming success before running commands | No evidence the claim is true | Run the command, read full output, then claim |
| Trusting results from a prior run | State may have changed since last execution | Always run fresh verification in the current session |
| Partial verification (tests but not lint) | Lint failures fail CI even when tests pass | Run all gates: tests, pnpm lint:fix, pnpm format |
| Using "should" or "probably" language | Implies assumption, not verification | Eliminate hedging; verify then state the fact |
| Skipping red-green-refactor cycle | Regression tests that always pass catch nothing | Verify test fails on revert before marking complete |
Before starting: Read .claude/context/memory/learnings.md
After completing:
.claude/context/memory/learnings.md.claude/context/memory/issues.md.claude/context/memory/decisions.mdASSUME INTERRUPTION: If it's not in memory, it didn't happen.
Weekly Installs
57
Repository
GitHub Stars
20
First Seen
Jan 27, 2026
Security Audits
Gen Agent Trust HubPassSocketPassSnykPass
Installed on
github-copilot56
gemini-cli55
kimi-cli54
amp54
codex54
opencode54
后端测试指南:API端点、业务逻辑与数据库测试最佳实践
11,800 周安装
学术评估工具 | ScholarEval框架:系统评估研究论文、文献综述与学术写作质量
70 周安装
豆包水印去除工具 - Python脚本一键移除豆包AI生成水印,支持批量处理
45 周安装
Salesforce B2C Commerce Page Designer开发指南:创建自定义页面与组件类型
70 周安装
AI研究综合器:多源信息整合与矛盾解决工具,生成结构化研究报告
70 周安装
API设计最佳实践指南:RESTful原则、命名规范与错误处理
70 周安装
Atlassian管理员专家:Jira/Confluence云与数据中心管理、用户安全与集成配置
71 周安装