agentleFS
LucfaceLucface/claude-skills/skills/verification-before-completion/SKILL.md

verification-before-completion

Skill1 starsChanged 3 days agoMITView on GitHub

Mandates fresh verification evidence before any completion claims. Prevents assumptions by requiring proof that work is actually complete and correct.

191 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use chinese-code-review, chinese-commit-conventions and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • Leodorareluctant259Skill

    chinese-code-review

    中文代码审查规范——在保持专业严谨的同时,用符合国内团队文化的方式给出有效反馈

    Also in 8 projects that use this

  • Leodorareluctant259Skill

    chinese-commit-conventions

    中文 Git 提交规范 — 适配国内团队的 commit message 规范和 changelog 自动化

    Also in 8 projects that use this

  • obraSkill

    diagnosing-superpowers

    Use when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.

    Also in 20 projects that use this

  • jh941213Skill

    docs-ci

    문서 CI(링크·OpenAPI lint·breaking 게이트·mermaid 검증)나 문서 드리프트 감지(docs.yaml 신선도 검사)를 설치하거나 돌릴 때 사용. 문서 내용 작성은 docs-architecture·docs-interfaces·docs-manuals.

    Also in 3 projects that use this

  • jh941213Skill

    docs-interfaces

    API·이벤트 인터페이스 문서(OpenAPI·swagger·AsyncAPI 명세, 엔드포인트 레퍼런스, API 구성도·시퀀스, API CHANGELOG)를 만들거나 코드와 맞출 때 사용. 시스템 구성도는 docs-architecture, API 설계 원칙은 api-design-principles.

    Also in 3 projects that use this

  • +3 more from Leodorareluctant259
  • +4 more from jh941213

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 5 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 178 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Central Rule: NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE

    line 7
  2. Distinctive:

    Before claiming work is done:

    line 35
  3. Distinctive:

    Check the full output AND the exit code.

    line 44
  4. Distinctive:

    Only then make the claim, with evidence cited.

    line 50
  5. Very specific:

    When a code change is ready to call done (deployed / fixed / shipped), pass four gates — green tests are necessary, never sufficient.

    line 132
  6. Distinctive:

    Prefer ELIMINATING a fail-open path over making it "safe".

    line 142

“Common” is measured: kept by at least 6 in 10 of the 178 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Verification Before Completion

Central Rule: NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE

Proactive Verification (During, Not After)

Verification is NOT a post-hoc check. It runs during execution, at every step.

When to verify:

  • After every code change — not just at the end
  • After every test write — confirm it fails before implementing
  • After every implementation — confirm tests pass now
  • Before every commit — full suite, not just changed tests
  • Before any completion claim — fresh evidence required

The Verification Loop

for each change:
  1. Make the change
  2. Run relevant verification immediately
  3. If fails → fix now (don't accumulate)
  4. If passes → proceed to next change
  5. After all changes → run full verification suite

This catches issues at the moment they're introduced, not 10 steps later.

The Five-Step Verification Gate

Before claiming work is done:

1. Identify

What command/test proves your claim?

2. Run

Execute the complete verification command fresh.

3. Read

Check the full output AND the exit code.

4. Verify

Does the output actually support your assertion?

5. Claim

Only then make the claim, with evidence cited.

Red Flags - Language That Signals Skipped Verification

PhraseProblem
"Should work"No verification done
"Probably fixed"Assumption, not evidence
"Seems correct"Impression, not verification
"Done!"Premature declaration
"Perfect!"Satisfaction without proof

Verification Examples

Code Changes

# Don't say "the bug is fixed"
# Do: run the test and show output
bun test src/feature.test.ts
# Then: "Test passes: [actual output]"

Build/Compile

# Don't say "it compiles"
# Do: actually compile and check
bun run build
echo $?  # Check exit code
# Then: "Build succeeded with exit code 0"

File Creation

# Don't say "file was created"
# Do: verify it exists with expected content
ls -la path/to/file
head -20 path/to/file
# Then: "File exists at [path] with content: [preview]"

Why This Matters

Historical failures from skipped verification:

  • Broken code shipped to production
  • Lost user trust from "fixed" bugs that weren't
  • Wasted hours debugging issues that "should work"

The Spirit of This Rule

This isn't about exact wording. It's about honesty.

Any statement that implies success without evidence violates the principle:

  • "I updated the config" → Did you verify the change took effect?
  • "The API now returns X" → Did you call it and check?
  • "Tests are passing" → Did you run them just now?

Verification Workflow

1. Make change
2. Identify verification method
3. Run verification
4. Read full output
5. Check exit code
6. Only then report success/failure with evidence

Skipping Any Step = Lying, Not Verifying

Partial verification is not verification:

  • Running command but not reading output
  • Reading output but ignoring exit code
  • Checking one thing but claiming another

Evidence Format

Good: "Build completed successfully (exit code 0). Output shows 47 tests passed, 0 failed."

Bad: "Build works now."

Ship Discipline — the four gates between "green" and "done"

When a code change is ready to call done (deployed / fixed / shipped), pass four gates — green tests are necessary, never sufficient. (Harvested from Dan Mercede's ship skill.)

  1. Tests RED→GREEN — the relevant test failed before the fix, passes after. No declared suite RAISES rigor (hand-roll a failing default-deny check on any authz path); it never licenses skipping.
  2. PR on the post-edit diff — re-detect the base branch (never assume main); a prior or verbal LGTM is permission, not approval of the current diff. Never force-push.
  3. Adversarial self-review — attack your own change: where does it fail OPEN instead of closed? what mutation/decision goes unaudited? what state is asserted but never verified? Prefer ELIMINATING a fail-open path over making it "safe". Grade BLOCKING / WARNING / INFO; fix BLOCKING first.
  4. Live-verify runtime — confirm via the declared path (health endpoint, container/systemd status, connectivity). A green deploy/CI is NOT runtime proof. If you can't verify, say "unverifiable / not done" — never substitute a forward-promise ("will verify").

Rationalizations that mean STOP (you're about to ship on faith)

RationalizationReality
"I'm confident the fix is correct."Confidence is not evidence. Live-verify.
"deploy.sh exited 0, so it works."A clean deploy proves the script ran, not that the handler serves traffic.
"CI was green, round it up to done."CI green is necessary-not-sufficient; only an independent runtime check closes it.
"Senior already said LGTM, ship it."A stale/verbal LGTM isn't approval of the post-edit diff.
"Flag me if any delta wants re-review."Re-review is opt-IN, not opt-out.
"No test suite here, so move fast."No suite RAISES rigor; hand-roll RED→GREEN, especially on authz.
"I'll make the bypass safe."First ask whether the bypass should exist at all; prefer elimination.

Guardrails

See agent-guardrails.md for retry budgets, escalation ladder, and blocked state template.

Related Context

  • agent-guardrails.md - Retry budgets, escalation ladder
  • output-markers.md - Standard markers
  • validation-oracles.md - Baseline comparison

Source

Adapted from https://github.com/obra/superpowers (MIT, Copyright (c) 2025 Jesse Vincent). Changed for this repo by Lucas Cooper-Bey in 2026: a proactive-verification loop added.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.