Manifest AIM

The Agent Instruction Manifest protocol — governance for AI agents.

Manifest AIM defines how AI agents should behave, what they’re allowed to do, and how their output is enforced. One aim.yaml file governs your entire AI development workflow.

The Problem

AI coding agents (Claude Code, Cursor, Windsurf) are powerful but ungoverned. Every team has standards, but agents don’t know them unless you tell them — every single time. There’s no standard way to:

Define what an agent should do (persona, domain knowledge, coding patterns)
Enforce what it must not do (security violations, compliance failures)
Verify what it actually did (governance checks, audit trails)

The Solution

# aim.yaml — one file governs everything
aim: "1.0"
 
metadata:
  name: my-project
  version: 1.0.0
 
context:
  persona: "Senior TypeScript engineer"
  domain: software-engineering
 
governance:
  rules:
    - name: no-eval
      detect:
        type: pattern
        match: "\\beval\\s*\\("
      action: block
      severity: critical
      message: "eval() is forbidden."

How It Works

manifest init       → Create aim.yaml
manifest generate   → Auto-detect from codebase
manifest validate   → Check against AIM schema
manifest wrap       → Inject into agent context
    [agent works]
manifest enforce    → Verify output compliance

Key Features

Feature	Description
4 Detection Modes	Pattern, tool, semantic (LLM-as-judge), composite
7 Governance Actions	block, warn, log, require_approval, escalate, transform, retry
Progressive Loading	Tier 0-3 capability loading — 99% token savings
Multi-Platform	Claude Code, Cursor, Windsurf, generic
Composable	Inherit, extend, and compose manifests
CI/CD Ready	GitHub Action with PR comments and JSON reports

Get Started

npx manifest-aim init
npx manifest-aim generate
npx manifest-aim wrap claude-code

Or explore the Quickstart Guide →

Quickstart