For people building things
Give your AI the context your best teammate already has.
Developers and product managers spend hours re-explaining the product, the users, the architecture, and the decisions behind the work. A Mara Playbook gives AI that working context before the task begins.
Give coding assistants your conventions, architectural decisions, constraints, and definition of done—without rebuilding that context in every conversation.
Less prompting. Less correction. More useful work on the first pass.
engineering-playbook/
├── MY-CONTEXT.md
│ └── stack, services, ownership, constraints
├── PLAYBOOK.md
│ └── conventions, review rules, definition of done
├── EXAMPLES.md
│ └── past changes, tests, commit style
└── BOUNDARIES.md
│ └── never publish, send, deploy, or invent
Asked: “Plan the retry work for the ingest worker.”
Queue changes go in the worker package, not the route handler —
matches the boundary you set for request-path code.
Retries: capped exponential backoff, dead-letter after 5 attempts.
Done means: unit test on the backoff maths + a runbook note.
A representation of the exported files — not a command-line tool.
Where the files go
- Claude. Drop CLAUDE.md at your project root; every new session reads it first.
- Codex and agent CLIs. AGENTS.md sits alongside your code and is picked up automatically.
- Gemini. GEMINI.md carries the same context for Google's tools.
- Anything else. The Playbook is plain text. Paste the copy-ready block into any chat.
Nothing here connects to your repositories, deploys anything, or sends messages. Mara writes context; you decide what to do with it.