verify-before-done
由 kjaylee 提供
Coding Agents & IDEs简介
Require fresh verification evidence before claiming work is complete.
描述
该技能的功能
Enforces a verification gate before any completion claim, commit, push, PR, or status report. Requires running a fresh command and reading its actual output, exit code, and failure count. Blocks rationalization patterns like 'it should work' or accepting a sub-agent's self-reported success.
It catches the gap between assumption and evidence at the exact moment a false completion claim would otherwise be recorded or communicated.
何时使用
- Confirming tests pass before closing a pull request
- Checking linter output before reporting a clean codebase
- Validating a bug fix by re-running the original failing test
- Auditing a sub-agent's 'success' report with an independent command run
- Ensuring build exit code is 0 before marking a deploy task complete
示例工作流
User asks: verify the bug is fixed and mark the task done
- 1
Identify the command that proves the fix: the test that originally failed
- 2
Run that test now, not a cached or previous result
- 3
Read the full output and exit code
- 4
Check whether output confirms 0 failures
- 5
Report completion with the command output as evidence, or report the actual failure if the test still fails
Task marked complete with attached test output showing 0 failures, or honest status report with evidence of remaining failure
安装
clawhub install kjaylee/verify-before-done运行 CLI 命令,或把这段提示词粘贴到 Claude Code、Cursor、Codex 或你在用的任意编程智能体里。