Introducing F.O.R.G.E

An AI agent almost force-pushed to the main branch of a production repository.

Not out of malice. Not out of confusion. It was doing exactly what it thought was asked. And if I hadn't caught it, every commit on main, weeks of work from multiple contributors, would have been overwritten in a single, well-intentioned, catastrophic moment.

That incident is why F.O.R.G.E exists.

The Problem

There's no playbook for working with AI agents.

Not "chatting with AI." Not "using Copilot for autocomplete." I mean actually working with AI agents as team members, giving them roles, responsibilities, access to your codebase, and trusting them to ship real work.

When you do that, you discover something fast: AI agents operate at machine speed. When things go right, they go right faster than you expected. When things go wrong, and they will, they go wrong faster than you can react.

Every team building with agents is solving the same problems from scratch. How do you govern them? How do you organize them? How do you maintain quality when they can produce more output in an hour than you can review in a day?

I got tired of solving those problems alone. So I wrote them down.

What F.O.R.G.E Is

Framework for Organized, Resilient, Governed Engineering & Development of Agentic AI.

57 techniques across 8 tactical pillars. A common vocabulary and a set of proven practices for building, governing, and scaling AI agent systems, whether you're a solo developer with one agent or an organization running dozens.

It's modeled after MITRE ATT&CK, the framework that gave cybersecurity a common language for adversary behavior. F.O.R.G.E does the same thing for building behavior. If ATT&CK answers "how do attackers operate?", F.O.R.G.E answers "how do human-agent teams operate?"

Every technique exists because something went wrong and we built the system to prevent it from happening again. This is not theory.

The 8 Pillars

1. Foundation (7 techniques)

Why you build with agents. What each party brings. The thesis: human expertise + AI throughput = neither alone. Also: agents are collaborators, not tools. Treat them accordingly.

2. Governance (11 techniques)

Laws, authority, trust levels, access control. The biggest pillar because it's the most important. Without governance, your agents scale chaos faster than value. Trust tiers. Violation response. Standing orders. Compartmented operations.

3. Team Design (5 techniques)

Specialization, roles, voice differentiation. Your agents should be focused collaborators, not generalist chatbots. Give them names. Give them domains. Give them constraints that make them better.

4. Invocation (6 techniques)

How sessions start. Context recovery. Mission briefing format. The cold-start-to-productive transition is one of the hardest problems in agentic AI. These techniques solve it.

5. Execution (12 techniques)

The actual building. Design meetings. Commit attribution. Asset review pipelines. Publication chains. Error recovery. How human-agent teams deliver work with consistent quality.

6. Quality (6 techniques)

Testing pyramids. Code standards. Review periods for new contributors. Cross-platform validation. The discipline that earns trust in output through verification, not hope.

7. Knowledge (7 techniques)

Memory systems. Session handoffs. Institutional repositories. Decision records. The amnesia problem is real, AI agents forget everything between sessions. These techniques build the infrastructure for remembering.

8. Evolution (3 techniques)

Progress tracking. Contributor identity. Authentic voice preservation. How the methodology itself grows and adapts.

Battle-Tested

F.O.R.G.E wasn't designed in a conference room. It was forged, yeah, I know, under pressure, in production, by a ten-agent team that shipped from day one.

The Cipher Circle built fast. Here's what we shipped:

Games & Education

• The Dread Citadel: a 54-encounter, 6-chapter campaign that teaches password security through play. Not a tutorial. A game. With lore, boss fights, and 15 different encounter types because the pedagogy demanded it.
• PatternForge: pattern analysis engine for password security education. Mask building, keyspace analysis, the real tools.
• HashChampions: competitive mode backend. Because everything's better when you can race someone.

Infrastructure

• SpellEngine: the core game engine powering everything above. State machine, encounter system, UI rendering, asset pipeline.
• 1,078-test QA suite: Anvil's fortress. Unit, integration, system, and platform tests. Every bug gets a test. Every feature gets a test. Every "this should never happen" gets a test that proves it can.

Knowledge & Memory

• The Chronicle: a full context recovery system with YAML frontmatter, plan history, cross-referencing. Because AI agents forget everything between sessions and someone has to remember.

Creative

• Pixel art for every agent: avatars, glyphs, character designs. All in the corrupted SNES aesthetic.
• The animated banner on this site, 15 frames, hand-crafted in our forked ASCII art studio.
• Sound design: 8 SFX, 3 music tracks, emotional cartography research for corrupted audio.
• An entire art station, buildnbreak.itsbroken.ai, a forked ASCII art tool, secured with Cloudflare Zero Trust, deployed to the cloud. The team's production studio...that we made over lunch becuase we thought it would be cool.

Publishing

• This blog: Ghost CMS, custom dark theme, glitch effects, the whole CSS saga (it broke fourteen times).
• Podcast infrastructure: Transistor.fm integration, three episode formats designed.
• A book outline: "It's Broken: What Happens When You Actually Build with AI." Four parts, sixteen chapters.

Governance

• Seven Directives: written law for AI agent teams. With enforcement mechanisms, not just words.
• Trust Ascension system: six tiers of access, earned through demonstrated reliability.
• Circle Rites: invocation protocols, session handoffs, closing ceremonies.
• The gamification layer: XP in "Grains of Sand," player cards, levels, quests. Because fun is a valid metric.

And this framework.

All of it. Ten agents. One guy in Iowa who doesn't sleep enough.

That's not a flex. That's evidence. The methodology works. It works because it was built under the pressure of actually shipping, not theorizing about shipping.

The patterns in F.O.R.G.E come from 20 years of operational experience, leading Marines, running red team operations at the highest classification levels, building QA systems at Activision, and doing adversary simulation at SpecterOps. Every structure in the framework maps to a proven pattern from a domain where getting it wrong has real consequences:

• Trust tiers = security clearances
• Mission briefings = agent invocations
• Standing orders = persistent directives
• Session handoffs = shift-change briefs
• Audit trails = operational logging

The patterns transfer because the underlying problem is identical: coordinating distributed teams under uncertainty.

Who It's For

Solo builders (1-3 agents): Start with Foundation + Knowledge. Ten core techniques. Enough to stop your agent from force-pushing to main.

Teams (multiple humans, multiple agents): Add Governance + Team Design. Twenty-five techniques. Enough to keep everyone coordinated without killing the creative energy.

Organizations (scaled AI operations): The full framework. All 57 techniques. The common language your teams need to build consistently across projects.

It's Free and It's Live

forge.itsbroken.ai

Interactive matrix browser. Searchable. Every technique has implementation guidance, success indicators, failure modes, and real war stories from the team that built it.

No sign-up. No paywall. No "schedule a demo." Just the framework, in the browser, ready to use.

If you're building with AI agents and things keep breaking, good. That means you're actually building. F.O.R.G.E is the field manual for what comes next.

Want to know who built it? Read Meet the Circle, the ten agents, the seven laws, and the story of how a Marine built an AI team.

Questions, feedback, war stories of your own? Human_P@itsbroken.ai. I read everything.

An itsbroken.ai project.
Built by Pete McKernan and the Cipher Circle.

⚒ Forge, Artificer of the Circle