What it does

Brigade is a file-based control layer for people who run several coding agents across tools, repositories and sessions. Its work ledger is machine-local JSON, nothing runs until a command or your own scheduler starts it, and it needs no background daemon. It lists unblocked tasks and uses atomic claims so two agents cannot take the same item. A check run through Brigade records the command, its real exit code and the Git state as a receipt, and declared wall-clock and worker-dispatch limits are enforced when you set them. Shared setup is handled by dry-run-first projection: MCP servers, tools and skills are previewed before Brigade writes them into an agent's own configuration. Notes handed between agents are checked and routed by a memory owner, and recall at session start is capped. Skills are scored from captured receipts and can be promoted or rolled back. The site is explicit about limits. Brigade is not a hosted memory service, not an automatic release bot and not a full autonomous fleet runtime, and it does not push to GitHub or publish packages on its own. Install it with pipx or uv; the README names a stable release and a beta channel.

Capabilities

Ready-work listing with atomic claims so agents do not pick the same task
Verification receipts with the command, exit code and Git state
Dry-run-first projection of MCP servers, tools and skills into agent configs
Linted handoffs and capped session-start memory recall
Outcome-based skill scoring with promotion and rollback
Optional fleet hub and read-only views of agent activity

When to use it

You run more than one coding agent and want one reviewed source for their shared tools and skills
You want proof of what a check actually ran and how it exited
You prefer local files and no daemon over a hosted service

Alternatives to explore

Beads
Rulesync
add-mcp