Verification Before Completion

This is the discipline that stops you from typing "done" before you've actually verified anything works. It forces a specific sequence: identify the command that would prove your claim, run it fresh and complete, read the actual output, then state your claim with evidence or admit it's not ready. The skill includes structured slots for what you checked, what you didn't, residual risk, and a confidence grade. It's surprisingly hard to follow when you're tired and just want the task over, which is exactly when you need it most. The paranoia extends to goal closure, workspace integrity checks, and explicit accounting of covered versus uncovered scope. If you've ever pushed code assuming the tests passed, this is the guardrail you were missing.

GitHub stars: 492
last updated: Jun 11, 2026

View on GitHub

Install

$ npx -y skills add ganyuanran/aegis --skill verification-before-completion --agent claude-code

Installs into .claude/skills of the current project.

Files

If `activation_mode = "explicit"` or `AEGIS_ACTIVATION_MODE=explicit` is visible and the request names neither Aegis nor this skill, return to the fast path without checklist/ceremony; otherwise continue.

Execute

Before success claim: classify destructive permission; read sufficient current evidence and run a fresh falsifying check when it is missing or invalidated; read the complete result/scope; select L0/L1/L2. Downgrade for partial, stale, failing, or narrower evidence; never claim complete then verify later.

This pack grants no GateDecision, PolicySnapshot, evidence sufficiency, requirement acceptance, or completion authority.