Free tool · AGENTS.md
AGENTS.md memory rules generator
Give any coding agent a memory protocol: recall before planning, claim tasks atomically, record decisions, and hand off context. Paste the block into AGENTS.md or CLAUDE.md.
Install locally
npx @awareness.market/setupWrites the rules and MCP config for every IDE it detects. Local-first; no account.
Agent / IDE
Paste this into AGENTS.md, CLAUDE.md, or your editor's rules file. The setup CLI does it automatically.
Generated rules
# my-project — agent memory rules
## Session start
1. Load memory before planning: `awareness_init`, or `awareness_recall` with the current task.
2. Check the shared task board: `awareness_lookup(type="tasks", status="open")`.
## Before starting work
- Another agent or IDE may touch this repo: claim the task first.
`awareness_record(action="update_task", task_id, status="claimed", expect_status="open", agent_role="claude_code")`
- If the claim returns `conflict`, pick different work. Never duplicate the task.
## While working
- Record decisions, constraints, pitfalls, and open questions. Never record secrets.
- On a long task, leave a handoff before stopping: goal / files / decisions / open questions.
## Before ending
- Record what changed and the next concrete step.
- Mark the task done; a reviewer attests with `verified_by`.Why a rules file is not enough
AGENTS.md tells an agent how to behave. It does not tell it what the last session decided, which approach failed, or what is still open. The generated block closes that loop by making the agent load memory before planning and write it back before stopping.
FAQ
What does this generator produce?
A ready-to-paste AGENTS.md / CLAUDE.md block: load memory at session start, claim tasks atomically, record decisions, and hand off context. It adapts to your project name, agent, and whether you run parallel agents.
Do I still need Awareness installed?
Yes. The rules call the awareness tools. Run npx @awareness.market/setup to install the local daemon and write the MCP config, or add the MCP entry manually from our MCP config generator.
Is this enough for multi-agent work?
The block covers the protocol, but the hard guarantees come from the daemon: atomic claims, lease reclaim, handoffs, and attestation. The rules just make agents use them.
Does it work for any agent?
Any agent that reads AGENTS.md, CLAUDE.md, or an equivalent rules file and can call MCP tools. That includes Claude Code, Cursor, Windsurf, Cline, Codex, and OpenCode.
Need the transport config? Use the MCP config generator. Per-IDE guides: Claude Code · Cursor.
Give your agent a memory that compounds.
Local-first, no account, no credit card. Start in 30 seconds and keep the data on your machine.