ANSS: A Specification Standard Built for AI Agents, Not Just Humans

TL;DR: Classic spec formats weren't designed for AI agents. ANSS is an open-source standard that gives both humans and coding agents a shared, unambiguous language. Cut my iteration count from 5–7 to 2–3 on real projects. Free on GitHub.

---

For a human developer, a good spec communicates intent. For an AI agent, a good spec eliminates decision space. These are different documents. ANSS is built to be both.

Three-layer architecture

[D] Domain        — WHAT (business requirements, user stories)
[E] Engineering   — HOW (architecture, stack, infrastructure)
[A] Agent         — INSTRUCTIONS (invariants, principles, rules) ← read first

Invariants

INV-001: No external npm packages
Cannot:  add require() of npm modules
Reason:  app must run without npm install
Check:   no node_modules imports in server.js

The Check field: a verifiable condition, not a comment.

Architectural Principles

Code	Principle
AP-001	Simple > Clever
AP-002	Monolith First
AP-004	Human Override Always
AP-006	Fail Loud
AP-010	Reversibility > Efficiency

Agent Review

Hard rule: > 3 problems found → stop and report. Never guess past this threshold.

Change Specification

Current State / Desired State / What NOT to Change / Impact / Rollback

---

git clone https://github.com/Kholomyanskiy/anss-standard
# Fill ANSS-Core-Template.en.md
# Tell agent: "Run Agent Review (Section 14) before writing any code."

Artem Kholomyanskiy (Artem Kholomianskyi) — AI Implementation Consultant, EVAI Consulting