Procedures
belong to
workflows
Open framework • AI-agnostic • Portable
The Standard for AI Engineering
A soft harness for capable models.
The smallest surrounding system that lets an agent do routine work inside your constraints, without asking first. It adapts to you, not you to it.
git clone https://github.com/AlexeyPlatkovsky/agent-manifestoModel-agnostic. Provider-agnostic. Future-proof.
- Claude
- Codex
- GPT
- Gemini
- Llama
- Grok
- any model
00 — OVERVIEW
What the model cannot infer
Write down the things a model has no way to know, once, where it already looks.
Models can read code, search, and reason. They cannot know your business, your conventions, or where your authority boundaries sit.
Agent Manifesto supplies the smallest surrounding system for that gap: context, policies, procedures, and boundaries for Claude Code, Codex, and other capable agents.
It is a specification plus a small plugin. You write down the things a model has no way to infer, once, in places it knows how to look. Then you keep it small: the manifesto is maintained by subtraction as well as addition, because an instruction nobody follows costs more than it saves.
If you write code
Your review gates, branch rules, and house conventions stated once, then applied on every session without you restating them.
If you are learning
A worked example of how much structure an AI actually needs to be useful, and how little you can get away with.
If you do not write code
A way to say what the agent may decide on its own and what has to come back to you, in plain sentences rather than settings.
01 — THE PRINCIPLES
Every concern has one home
The manifesto is a mapping. Put each kind of thing where it belongs, and the harness holds.
Policies
belong to
skills
Responsibilities
belong to
agents
Data shapes
belong to
contracts
Verification
belongs to
evidence
Reasoning
belongs to
the model
02 — INSTALL
Add the marketplace, install, inspect
Ships as a plugin from GitHub sources. Three skills, one agent, ~189 tokens of discovery metadata.
Claude Code
claude plugin marketplace add ./claude plugin install agent-manifesto@agent-manifesto --scope local -yclaude plugin details agent-manifesto # component inventory and token cost
Codex
codex plugin marketplace add .codex plugin add agent-manifesto@agent-manifestocodex plugin list
03 — WHAT'S IN THE PLUGIN
Small surface, five moves
Enough structure to route work and keep it honest — nothing that has to justify its weight twice.
Agent
- instruction-evaluatorValidates the landscape against its schema. Claude Code only.
Skills
- ai-landscapeRoutes a request to the right workflow.
- brainstormSurfaces meaningful alternatives before a decision.
- documentation-maintenanceKeeps docs in step with the change.
Workflows
- setupEstablish instructions for a project.
- reviewAudit instructions for relevance.
- extendAdd a skill, workflow, agent, or rule.
- adoptBring an external tool inside the landscape.
- retirePreserve the essential facts, then remove.
04 — ECOSYSTEM
Built on the same rig
Companion tools that assume the same homes for the same concerns.
Agent Manifesto
The versioned spec and the plugin. What to do, what not to, and why — maintained by subtraction as well as addition.
Read the manifestoPlayForge
Playwright + TypeScript + AI pipelines. Deterministic browser tests that produce evidence, not vibes.
PlayForgeTaskPilot
Local-first issue tracking for AI. Item lifecycle, hierarchy, and status that agents can read and write.
TaskPilot