Agent-Ready Repo
Every vibe coder hits the same wall: after a few sessions, the AI agent starts contradicting itself, breaking things it already fixed, or going in circles. The context window fills up and the agent loses the plot.This pack is the system that fixes that. It's 6 files you drop into any project that give your AI agent persistent memory, non-negotiable rules, and a clear map of your codebase — so it stays coherent from session one to session one hundred.Built and battle-tested across real production projects. Not theory.What's IncludedFileWhat It DoesCLAUDE.mdThe agent's entry point. Tells it what to read first, enforces the Council Protocol, and defines quality gates. Drop this in your project root._ai/RULES.mdNon-negotiable rules: threading, core stability, platform constraints, data mutation. Violations cause crashes. Agent reads this before touching anything._ai/ARCHITECTURE.mdThe system map. Every key file, the data flow, entry points, DB schema, project structure. Agents that know the map make surgical changes instead of wild guesses._ai/DECISIONS.mdWhy the code is the way it is. Prevents agents from 'fixing' intentional decisions. The most underrated file in the pack.ai/SESSIONHANDOFF.mdWritten by the agent at the end of every session. Read at the start of the next. The only memory that survives between sessions._ai/BUILD.mdBuild commands, test instructions, environment setup. Loaded only when building or testing — keeps the main context lean.The Council ProtocolThe most important feature. Baked into CLAUDE.md. Required before any risky change:1. Propose approachAgent describes what it plans to do and why.2. Critique itAgent lists what could go wrong. Forces it to think before acting.3. ReviseAgent updates the approach based on its own critique.4. ImplementOnly now does it write code.Skip for single-file, single-function fixes. Mandatory for anything that touches more than 2 files, core modules, or the database.How To Use This PackStep 1 — Copy the filesDrop CLAUDE.md into your project root. Create an _ai/ folder and put the other 5 files inside it.your-project/ CLAUDE.md _ai/ RULES.md ARCHITECTURE.md DECISIONS.md SESSION_HANDOFF.md BUILD.mdStep 2 — Fill in your project detailsEvery file has clear [PLACEHOLDER] sections. Fill them in once per project. Takes 20–30 minutes. Saves hours of agent confusion.Key things to fill in: Project name and one-sentence description in CLAUDE.md Threading rules specific to your language/framework in RULES.md Your actual file paths and data flow in ARCHITECTURE.md Any intentional decisions that look 'wrong' in DECISIONS.md Your build commands and test setup in BUILD.md Step 3 — Tell the agent to read it firstStart every session with:Read CLAUDE.md and all files it references before doing anything.The agent will load the full context and be ready. Works with Claude Code, Cursor, Windsurf, and any other agent that reads markdown. Step 4 — Let the agent handle SESSION_HANDOFF.md At the end of every session, the agent writes what it did, what it changed, and what's next. Next session, it reads that first. You never have to re-explain context.Compatible AgentsTested with: Claude Code (claude.ai/code) — primary target Cursor — works via .cursorrules or direct file reference Windsurf — works via cascade context Any agent that reads markdown files before acting Why This WorksAI agents don't have memory between sessions. They start fresh every time. Without a system like this, every session is the agent learning your codebase from scratch — making the same mistakes, asking the same questions, breaking the same things.This pack gives the agent a persistent brain. RULES.md is its discipline. ARCHITECTURE.md is its map. DECISIONS.md is its institutional knowledge. SESSION_HANDOFF.md is its diary. CLAUDE.md is the protocol that ties it all together.The result: agents that make surgical changes, don't break stable code, and improve with every session instead of regressing.
Get it → abesoftt.gumroad.com