一键导入
eng-verification
Mandatory checklist before claiming work is complete—run the evidence-producing command, inspect output, and only then state a result.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Mandatory checklist before claiming work is complete—run the evidence-producing command, inspect output, and only then state a result.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Model the end-to-end system, constraints, and failure modes before touching code; use whenever a change spans multiple layers or unclear requirements.
Decompose work into safe, observable increments with feature flags, rollbacks, and communication baked in.
Design every change with traceability, diagnostics, and fast incident triage in mind across mobile, web, and web3 stacks.
Guard latency, memory, battery, bandwidth, and gas/compute budgets by measuring before and after every change.
Deliver changes with clear communication, reviewer-ready context, and follow-up accountability.
Enforce reproducible testing, automated verification, and evidence-driven sign-off before merging any change.
| name | eng-verification |
| description | Mandatory checklist before claiming work is complete—run the evidence-producing command, inspect output, and only then state a result. |
Claiming success without evidence is dishonesty, not efficiency. Every status report, “done,” or “tests pass” statement requires fresh verification.
Iron Law
NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE
If you didn’t run the verifying command in this message, you cannot claim it passes.
1. IDENTIFY: Which command proves the claim?
2. RUN: Execute the full command now (fresh run).
3. READ: Inspect entire output & exit code.
4. VERIFY: Does output prove the claim?
- If NO → report actual status with evidence
- If YES → state claim WITH evidence (command + summary)
5. ONLY THEN: Make the claim
Skipping any step = lying.
| Claim | Required | Not Enough |
|---|---|---|
| “Tests pass” | Latest test command output showing 0 failures | Previous run, “should pass” |
| “Lint clean” | Linter command exit 0 now | Partial subset, inference |
| “Build works” | Build command exit 0 | Linter pass |
| “Bug fixed” | Reproduction test now passes | Code diff only |
| “Regression test exists” | Red→Green verified cycle | Test created but never failed |
| “Agent succeeded” | Inspect VCS diff + verify outputs | Agent success message |
| Excuse | Reality |
|---|---|
| “Should work now” | Run the command |
| “I’m confident” | Confidence ≠ evidence |
| “Just this once” | No exceptions |
| “Agent said success” | Verify independently |
| “Partial check is fine” | Partial proves nothing |
| “I’m exhausted” | Exhaustion isn’t evidence |
X/Y passing output.Always before:
Evidence precedes claims. Run the command, read the output, then speak. Anything else breaks trust.