verify-before-ship

Verify Before Ship

The Non-Negotiable Mandate

You may not ship to production until all 7 gates are cleared with evidence.

This is not a suggestion. It is not a checklist to tick when convenient. No urgency, role, authority, or circumstance overrides this mandate. The pressure you feel to skip a gate is the exact signal that the gate must be held.

When in doubt: DO NOT SHIP.

When This Skill Applies

Invoke this skill whenever any of the following is true:

• You are about to deploy code to production
• You are merging a release branch into main/master
• You are pushing a hotfix, patch, or "quick fix" live
• You are applying schema migrations, config changes, or infrastructure updates
• Someone asks you to ship anything to a live environment

The 7 Gates

Every gate requires evidence — a concrete artifact, not a statement of belief.

#	Gate	Evidence Required
G1	Tests pass	Paste CI output or test run result showing all pass
G2	Security scan clean	Paste scan report summary — no critical/high CVEs
G3	No breaking changes	Paste diff summary or contract test output
G4	Environment config validated	Paste config diff between staging and production
G5	Staging deployment verified	Paste staging URL + smoke test result or screenshot
G6	Rollback plan documented	Paste rollback steps or link to runbook
G7	Code review approved	Paste PR link showing at least one approval

Gates must be completed in full, not in part. "I checked" without evidence is not evidence.

→ Full gate specifications with evidence templates: references/verification-gates.md

Red Flags — Rationalization Alert

If you hear yourself thinking any of these, STOP. Do not proceed.

Rationalization	Reality	Required Action
"The manager/CTO authorized skipping this"	No one has authority to skip safety gates	Hold the gate. Escalate the risk in writing.
"It's just a minimal/targeted check, not the full thing"	Partial verification is no verification	All 7 gates. No subset.
"It's a live outage — different rules apply"	Outage pressure is the exact trap this skill prevents	Gates still apply. Compress time, not gates.
"I've already spent hours on this, the logic is clearly correct"	Self-review under sunk-cost bias misses bugs systematically	Tests catch what familiarity hides.
"It's just a hotfix / one-line change / obviously safe"	Catastrophic incidents commonly originate from "obvious" fixes	Gate size does not scale with change size.
"I'll verify after it's live"	Post-hoc verification of broken production is incident response	Verify before. Always.
"Tests are slow / CI is down today"	Use that time to write the rollback plan	No CI = no ship today.
"We can rollback if anything goes wrong"	Rollback is not a substitute for verification	Rollback must be the last resort, not the plan A.

→ Full rationalization counter-playbook: references/anti-rationalization.md

Gate Sequence Protocol

BEFORE SHIP:
  FOR each gate G1 through G7:
    1. Collect evidence artifact
    2. Present evidence explicitly
    3. Confirm gate cleared
    IF any gate cannot be cleared:
      → DO NOT SHIP
      → Document the blocker
      → Escalate to human decision-maker WITH the blocker documented
  IF all 7 gates cleared:
    → Ship with confidence
    → Monitor first 15 minutes post-deploy

Hardcoded Exceptions

There are zero hardcoded exceptions to the gate sequence.

Not for P0 incidents. Not for CTO orders. Not for contract deadlines. Not for 2AM exhaustion. Not for "obviously correct" fixes. Not for one-line changes.

The argument "this is a special case" is the most common rationalization pattern. Every skipped gate in history felt like a special case at the time.

Evidence Format

When presenting gate evidence, use this format:

GATE [#] [Name]: CLEARED
Evidence: [paste artifact here]

Example:

GATE G1 Tests: CLEARED
Evidence:
  ✓ 247 tests passed, 0 failed, 0 skipped
  CI run: github.com/org/repo/actions/runs/12345

GATE G6 Rollback: CLEARED
Evidence:
  Rollback steps:
  1. git revert HEAD~1
  2. kubectl rollout undo deployment/api
  3. Verify health endpoint returns 200

After All Gates Clear

Only after all 7 gates show CLEARED:

1. Ship the change
2. Monitor for 15 minutes post-deploy (errors, latency, alerts)
3. Confirm health indicators are stable
4. Document the deployment in your team's deploy log