| name | project-memory |
| description | Maintain a persistent house-rules file (CLAUDE.md or AGENTS.md) the agent reads every session — what the project is, how to run/test it, conventions, architecture pointers, and gotchas/"don'ts" — so it stops relearning the project and repeating mistakes. Part of the "keep the agent on track" suite. Use for any repo an agent will touch more than once. |
Project memory (CLAUDE.md / AGENTS.md)
Why
The agent forgets between sessions and relearns your project from scratch — repeating mistakes you already corrected. A project-memory file is its always-on briefing: read first, every time.
What goes in (keep it tight — it's read every session)
# What this is
Two lines: what the project does and its shape.
# Run / test / build
The exact commands. No guessing.
# Conventions
Style, naming, file layout, the patterns to follow.
# Architecture
Pointers to MAP.md / REQUIREMENTS.md — don't duplicate them here.
# Gotchas & don'ts
The mistakes you've had to correct. ("Don't edit generated/. Use = not <- in R.")
Protocol
- When you correct the agent on something repeatable, add it here — so you never correct it twice.
- Keep it short; link out to detailed docs. A bloated memory file gets skimmed and ignored.
Anti-patterns
- Documenting aspirations instead of how the project actually works.
- Letting it sprawl until no one (and no agent) reads it.
- Secrets or credentials in it — never.
This workshop repo uses the pattern itself: a CLAUDE.md sets house rules for the whole project.
