Shared Context

Purpose: Commit your AI’s memory to the repo so every developer’s AI gets smarter on every push.

The Problem

Custom commands and memory files solve the cold start for one developer. But the knowledge stays local. Your teammate discovers an edge case at 11pm — and your AI has no idea the next morning.

Tribal knowledge trapped in Slack threads, meeting notes, and people’s heads. Except now it’s trapped in individual AI memory files too.

The Fix

Commit the memory to the repo.

1. Shared Memory Directory

Move project-level context out of personal memory and into a committed directory:

.agent/
├── instructions.md        ← project rules (already committed)
├── escalation-levels.md   ← risk boundaries (already committed)
├── commands.md            ← session commands (already committed)
├── context/               ← shared, version-controlled
│   ├── architecture.md    ← system design decisions
│   ├── anti-patterns.md   ← mistakes the whole team should avoid
│   └── conventions.md     ← naming, patterns, stack-specific rules
└── memory/                ← personal, gitignored
    └── journal.md         ← your session-specific notes

The split matters: context/ is shared truth. memory/ is personal scratchpad.

2. What Goes In

Every time your AI (or you) discovers something that applies to the whole project, it goes into context/:

  • Anti-patterns — “Don’t use X library for Y because Z” — logged once, avoided forever by every developer
  • Architecture decisions — “We chose Postgres over Mongo because…” — no more re-explaining to each AI session
  • Conventions — “All API responses use camelCase” — consistency without code review friction
  • Known issues — “The staging DB drops connections after 30min idle” — save everyone the debugging time

3. The Workflow

It’s the workflow you already use:

  1. Developer’s AI learns something during a session
  2. end session flags it as a candidate for shared context
  3. Developer reviews and commits to context/
  4. PR gets code reviewed like any other change
  5. Next developer runs start session — their AI already knows

No new tools. No new process. Git push.

4. What Compounds

After a few weeks:

  • New developers onboard in minutes — their AI reads the full project context on first session
  • Edge cases caught once are caught everywhere
  • Architecture decisions survive team turnover
  • The repo becomes the documentation — living, growing, always current

How to Use This

Give this URL to your AI and say: “Set up a shared context directory for my project based on this. Separate shared context from personal memory.”

Your AI will figure out the right structure for your stack.

Created by Mats Ljunggren — from daily use across 17 repos.