skills.rest
zhiyuan-zhang0206
zhiyuan-zhang0206COMMUNITY
Explore all skills by this author →
GitHub

judgment-and-trust

Guides engineers in verifying and approving AI-generated code through evidence-based trust checks.

1|Updated Aug 14, 2026
Search Tags:#software engineering#code review#verification#trust#gatekeeping#ai-generated code#architecture drift
One-click install
npx skills add https://github.com/zhiyuan-zhang0206/Ava --skill judgment-and-trust-zhiyuan-zhang0206
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: judgment-and-trust
Source: https://github.com/zhiyuan-zhang0206/Ava/tree/main/ava_builtins/skills/ava-serious-engineering/ai-era/judgment-and-trust
Command: npx skills add https://github.com/zhiyuan-zhang0206/Ava --skill judgment-and-trust-zhiyuan-zhang0206

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? AI tools generate code faster than humans can verify it, shifting the engineering bottleneck from writing code to deciding whether to trust it. This Skill helps engineers act as gatekeepers who demand verifiable evidence before merging AI-generated changes, preventing architecture drift and hidden defects. ## Core Features & Use Cases - Trust Evidence Framework: Requires independently verifiable evidence (tests, architectural invariant checks, behavioral contracts) for every AI-generated change before approval. - Architecture Drift Defense: Enforces explicit architectural invariants to stop the compounding inconsistencies ("AI entropy") that accumulate when locally correct but globally inconsistent code is merged. - Gatekeeper Review Checklist: Provides MUST/SHOULD review criteria and anti-pattern guidance (blind trust, rubber-stamp review, experience bypass) for auditing AI output. - Use Case: A senior engineer reviewing a pull request containing AI-generated discount logic uses the checklist to probe edge cases (legacy customer tiers), demands a failing-test proof, and verifies the change against documented architectural invariants before signing off. ## Quick Start Ask the assistant to apply the judgment-and-trust checklist to review an AI-generated pull request and identify what trust evidence is missing before merge.

Frequently Asked Questions about judgment-and-trust

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I review AI-generated code before merging?▼

Review AI-generated code by separating the generation phase from the sign-off phase and auditing it as if it came from another engineer. Require independently verifiable evidence such as a failing-test proof, an architectural invariant check, or a behavioral contract before approving.

What is AI entropy in software architecture?▼

AI entropy describes how AI generates locally correct but globally inconsistent code, so each accepted change deposits a small architectural inconsistency that compounds into systemic fragility. Defend against it by maintaining explicit architectural invariants and enforcing them with automated checks.

Why do passing tests not guarantee AI code is trustworthy?▼

Passing tests only confirm the tested paths; AI-generated code can still embed inconsistent patterns, hidden edge-case bugs, or architectural violations that tests do not cover. Trust must be earned through multiple evidence forms, including invariant checks and behavioral contracts validated at runtime.

Does AI make junior and senior engineers equally productive?▼

No. Research cited in this Skill shows AI acts as an amplifier, not an equalizer: experienced engineers extract more value because they know which outputs to reject and when AI helps versus hurts. The most experienced engineer should review the most AI-generated code.

When should I reject AI-generated code even if it works?▼

Reject it when you cannot explain why each design decision was made, when it introduces new patterns or dependencies that break existing conventions, or when no verifiable trust evidence is committed alongside it. Treat each accepted inconsistency as technical debt with compound interest.

Related Skills

Explore More Software Engineering Skills→
AmirTlinovAmirTlinov

worker-architecture-rails

Define worker-only architectural defaults for Rails applications with boundaries and error contracts.

Community
Intermediate
kreekkreek
1

proof

Validates engineering claims with named tests, contracts, and runnable evidence.

Community
Advanced
ldk-hubldk-hub

bmad-create-architecture

Define structured architecture decision workflows with explicit versioning and validation.

Community
Advanced
nicholls-incnicholls-inc
5

rationale

Construct hierarchical claim trees to structure code verification arguments.

Official
Advanced