跳到正文
cccskills
免费在 GitHub 公开

verification-before-completion

Use before claiming work is complete, fixed, or passing — before committing, opening a PR, or handing off. Requires running the verification command in THIS turn and reading its output before any success claim.

中文简介正在准备中。当前显示原文说明。

查看安装方法

包含的文件(1)

  • SKILL.md4.0 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Verification Before Completion

Claiming work done without fresh verification is dishonesty, not efficiency. adversarial-verify is the what; this skill is the when — the gate you pass through right before any completion claim.

The Iron Law

NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE

If you have not run the verification command in this message, you cannot claim it passes. Not "should", not "probably", not "based on the diff".

The gate function

Before writing "done" / "fixed" / "green" / "ready to merge" — even in your own head:

  1. Identify — what exact command proves this claim?
  2. Run — execute it fresh, complete, in this turn.
  3. Read — full output, check exit code, count failures.
  4. Verify — does the output actually confirm the claim?
  5. Only then — make the claim, with the evidence attached.

Skip any step = you are lying to the user, not verifying.

Common false claims → what they actually need

ClaimRequiresNot sufficient
Tests passFresh test run, exit 0, 0 failures"should pass", previous run, "logic looks right"
Linter cleanLinter output, 0 errorsPartial check, extrapolating from unrelated files
Build succeedsBuild command, exit 0Linter passing, editor squiggles gone
Bug fixedReproduce original symptom, watch it not happenCode changed, "assumed" fixed
Regression test worksRed → green cycle verified (revert fix, watch test fail, restore, watch pass)Test passes once
Agent/subagent completedRead the VCS diff, verify claimed changes existAgent's own "success" report
Spec satisfiedLine-by-line checklist against the plan"Tests pass, phase complete"

Red flags — you are about to claim without verifying

  • Words like "should", "probably", "seems to", "looks good"
  • Satisfaction language ("Great!", "Perfect!", "Done!") before running the command
  • About to commit / push / open PR without a verification block in this turn
  • Trusting a subagent's own success report
  • "Just this once" thinking, or "I'm tired, close enough"
  • Partial verification (linter passed, so build must)

Rationalization prevention

ExcuseReality
"Should work now"RUN it.
"I'm confident"Confidence ≠ evidence.
"Linter passed"Linter ≠ compiler ≠ tests.
"The agent said success"Read the diff yourself.
"Partial check is enough"Partial proves nothing about the whole.
"Different words, so rule doesn't apply"Spirit over letter.

Patterns

Tests

  • Run the test command. See 34/34 pass. Then say "all tests pass".
  • Never: "should pass now".

Regression tests (real red-green)

  • Write test → run (pass) → revert fix → run (MUST FAIL) → restore fix → run (pass).
  • Never: "I've added a regression test" without the red-green cycle.

Build

  • Run the build. See exit 0. Then say "build passes".
  • Never: "linter passed, build should too".

Agent delegation

  • Subagent reports success → check the VCS diff → verify the claimed change is actually there → report actual state.
  • Never: paste the agent's report and treat it as truth.

When this fires

Always, before:

  • Any variation of success / completion / fixed / passing / green
  • Committing, opening a PR, marking a task done, handing off
  • Moving to the next task
  • Any positive statement about the work's state

Pair with

  • adversarial-verify — the 11 shortcuts agents take to fake "done"; run through the list, then run through this gate.
  • clean-commits — clean commits require verified content.
  • The verifier subagent — dispatch it; then verify its report against the diff (per the "Agent delegation" pattern above).

The bottom line

Run the command. Read the output. THEN claim the result. Non-negotiable.

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

a11y-pass

免费

Catch the accessibility failures that ship in almost every AI-built UI. Use after building any interactive component.

中文简介正在准备中。当前显示原文说明。

Archive228/loopkit7552026年7月15日 更新

Before compaction Loopkit extracts decisions into claude-decisions.json (machine-readable). Read it alongside claude-progress.txt at session start — prose is for humans, JSON is for the loop.

中文简介正在准备中。当前显示原文说明。

Archive228/loopkit7552026年7月15日 更新

Review a diff against the goal spec assuming the code is BROKEN. The reviewer that lives in the maker's head always agrees with itself — this pulls review into a hostile, separate pass. Invoke after every code change before marking work done.

中文简介正在准备中。当前显示原文说明。

Archive228/loopkit7552026年7月15日 更新

Verify that an endpoint checks ownership, not just authentication. Use on any handler that reads or mutates user data.

中文简介正在准备中。当前显示原文说明。

Archive228/loopkit7552026年7月15日 更新

Find the exact commit that introduced a bug. Use when something worked before and broke, and you don't know which change did it.

中文简介正在准备中。当前显示原文说明。

Archive228/loopkit7552026年7月15日 更新

Before picking new work, smoke-test the last "completed" feature. If it's broken, revert and re-open it before touching anything else. Kills the "looks shipped, isn't shipped" bug across sessions.

中文简介正在准备中。当前显示原文说明。

Archive228/loopkit7552026年7月15日 更新

查看 Archive228 的全部技能

报告此技能的问题