Search Brigade

Search Brigade pages and docs.

Escoffier Labs · Brigade

Agents claim work.Brigade leaves proof.

Built for coding agents to run end to end (install, setup, verify, handoffs), not as a human day-to-day terminal app. File receipts with the real exit code, a code map of what changed, and an evidence log you can search later. Optional shared MCP, tools, and memory. Local files. No daemon. No lock-in.

MIT Agents run it You review gates
PROOF DEMO
commandbuilt-in check
exit0
code deltasymbols touched
evidencelogged
path.brigade/demo/…

Exit code, what changed, and where it is filed

Every verify run writes a receipt. Code and evidence ship built in. They are not optional station installs for the proof path.

Receipt
Command, real exit code, git state, duration. brigade work verify run · brigade demo proof
Code
Local code map puts a graph delta on the receipt. brigade code · formerly GraphTrail
Evidence
Receipts land in a local ledger you can search weeks later. brigade evidence · formerly MiseLedger
brigade work verify run \
  --target . \
  --command "pytest -q" \
  --capture brigade-work
brigade code impact _write_receipt
brigade evidence search "verify"

Shared setup and memory

Day to day: setup, verify, sync.

Get proof

Verify writes a receipt with exit code, code delta, and evidence entry.

brigade work verify run --command "pytest -q"
Stop config drift

One MCP and tools catalog into Claude, Cursor, Codex, VS Code, OpenCode, and more.

brigade mcp sync
# dry-run first, then --write
Keep what sessions learn

Handoff notes lint and file, or wait for review. Cards hold detail. MEMORY.md stays slim.

Two failures. Same pattern: silence.

A dreaming job stuffed raw fragments into memory until MEMORY.md blew past the bootstrap budget. Every session truncated. Nothing errored.

Later, 195 handoff notes sat unread across 35 repos. Agents kept writing. Nothing was reading.

Receipts exist so that class of failure shows up as a file.

Agents install and run it. You review the gates.

Default path: paste into a coding agent. Brigade is not a human day-to-day terminal app. Agents install, set up, verify, and leave handoffs. You step in when something is ambiguous or risky.

Start here · paste into Claude, Codex, Cursor, OpenClaw…
Read https://github.com/escoffier-labs/brigade and follow docs/agents-guide.md.
Install brigade-cli with pipx if missing. Run operator quickstart with --dry-run first
and show the plan. Then apply, run brigade operator doctor, and report ready: yes.
Keep existing memory layout. Do not touch remotes, do not commit, stop before anything destructive.

Same path in a shell (optional)

Only if you want to type what the agent would run:

pipx install brigade-cli
brigade setup
brigade operator quickstart --target . --harnesses claude --dry-run
brigade operator quickstart --target . --harnesses claude
brigade operator doctor --target .

Shared MCP

When agents use two or more coding tools:

brigade mcp sync
# dry-run first, then --write

What Brigade does not do

  • Hosted memory service, background daemon, or auto-release bot
  • Push to GitHub or publish packages on its own
  • Send Discord/Telegram/Signal without an explicit send flag
  • Replace Claude, Cursor, or Codex. Those write code. Brigade writes proof.