Skip to main content

File-based Claude/Codex agent coordination with queues, driven roles, registered Ansible stand leases, and evidence gates

Project description

greatminds

File-based multi-agent coordination for agent fleets and task pipelines.

PyPI version Python versions License Docs

greatminds runs a fleet of Claude Code and OpenAI Codex agents on a shared filesystem-based finite state machine. Tasks flow through queues such as feature_inbox/, feature_plan/, feature_dev/, feature_test/, and verified/; a small coordd daemon nudges agents when input appears. There is no central broker and no database. Per-project setup writes coord.yaml; the per-user daemon can supervise multiple projects on one machine.

Quickstart

# install
pip install greatminds  # or: uv add greatminds

# bootstrap a project
mkdir -p /tmp/greatminds-demo
cd /tmp/greatminds-demo
greatminds setup --session myproject

greatminds supports two primary agent tools:

  • claude: Claude Code. Used for chat, loop, and driven roles. Setup writes .claude/settings.local.json and installs configured Claude marketplace plugins for Claude-hosted roles.
  • codex: OpenAI Codex. Used for chat and driven roles. Run codex login once for the machine account; generated files under coordination/.codex-home/<role>/ are role config sources, not auth homes.

Window modes in coord.yaml:

  • chat: a live tmux pane for an operator-facing conversation.
  • loop: a resident watchdog pane that wakes on its own timer.
  • staged: a tmux pane with the start command pre-typed, so the operator starts that role manually when needed.
  • driven: no live pane; coordd starts one Claude or Codex turn when work lands in the role's queue, inbox, or stand event stream.

Role-to-tool assignment lives in coord.yaml. Edit it after setup when you want different tools for different roles:

Role Default tool Default mode
ARCHITECT-PLANNER codex chat
MAINTAINER claude loop
LIVE-DEVELOPER claude staged
ARCHITECT-REVIEWER codex driven
DEVELOPER claude driven
UI-DEVELOPER claude driven
TECHNICAL-WRITER codex driven
TESTER claude driven
READER claude driven
EXPLORER codex driven
session: myproject
project_dir: /tmp/greatminds-demo
windows:
  - name: planner
    role: ARCHITECT-PLANNER
    tool: codex
    mode: chat
  - name: maintainer
    role: MAINTAINER
    tool: claude
    mode: loop
  - name: dev
    role: DEVELOPER
    tool: claude
    mode: driven
  - name: reviewer
    role: ARCHITECT-REVIEWER
    tool: codex
    mode: driven

Put project and stand variables in coordination/PROJECT.env. It is gitignored, sourced before agent launch, and passed to stand profiles as Ansible extra vars:

cat > coordination/PROJECT.env <<'EOF'
STAND_HOST=localhost
STAND_USER=violet
EOF

STAND_HOST=localhost is enough for a local smoke stand. For a remote stand, set STAND_HOST to an SSH config alias or a comma-separated list of aliases, and set STAND_USER to the remote account whose PATH should be used.

A stand is a singleton live environment. Agents lease it with greatminds stand lease; coordd deploys the leased worktree by running an Ansible playbook selected through coordination/stand-profiles.yaml. There is no global current profile. The current profile is chosen for each lease with greatminds stand lease --profile <name>, stored in coordination/.stand/state.yaml as active_lease.profile, and resolved through the registry to a YAML file under coordination/stand-profiles/.

Setup seeds reference profiles named full-deploy, vite-dev, and smoke-only. Add project profiles by adding registry entries and YAML files. The used_for and default_for values are machine-readable tokens from coordination/schema.yaml under stand_profile_registry; default_for tells roles which profile to choose for common lease purposes:

profiles:
  full-deploy:
    file: full-deploy.yaml
    purpose: Full deployed product validation on a stand.
    environment: stand
    used_for: [tester_validation, explorer_review, reviewer_validation]
    default_for: [feature_test, explorer, reviewer]

Inspect and validate the registry with:

greatminds stand profiles list
greatminds stand profiles doctor

For production, create an explicit registry entry with environment: production, requires_explicit_user_approval: true, and allowed_roles such as [ARCHITECT-REVIEWER, MAINTAINER]. Lease it only after approval:

greatminds stand lease \
  --task <task-id> \
  --profile production \
  --profile-approval USER_APPROVED

A minimal smoke playbook looks like this:

---
- name: register stand node
  hosts: localhost
  gather_facts: false
  tasks:
    - name: add configured stand host
      ansible.builtin.add_host:
        name: "{{ STAND_HOST | default('localhost') }}"
        groups: stand_nodes
        ansible_connection: >-
          {{ 'local' if (STAND_HOST | default('localhost')) == 'localhost'
             else 'ssh' }}

- name: smoke stand
  hosts: stand_nodes
  gather_facts: false
  tasks:
    - name: remote shell works
      ansible.builtin.command: /bin/true
      changed_when: false

Then start the fleet:

# install the per-project daemon
greatminds daemon install
greatminds daemon start

# launch agents
greatminds launch --target tmux
tmux a -t myproject

The windows defined in coord.yaml boot inside one tmux session. Roles with mode: driven do not keep live panes; coordd starts one Claude or Codex turn when their queue, inbox, or stand-state event changes.

Key Concepts

Documentation

Full documentation: https://veryviolet.github.io/greatminds/

Where to File Issues

Bugs in greatminds: https://github.com/veryviolet/greatminds/issues
Bugs in a project you use greatminds in: that project's issue tracker.

License

Apache-2.0. See LICENSE.

Project details


Release history Release notifications | RSS feed

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

greatminds-2.2.0.tar.gz (327.0 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

greatminds-2.2.0-py3-none-any.whl (382.8 kB view details)

Uploaded Python 3

File details

Details for the file greatminds-2.2.0.tar.gz.

File metadata

  • Download URL: greatminds-2.2.0.tar.gz
  • Upload date:
  • Size: 327.0 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for greatminds-2.2.0.tar.gz
Algorithm Hash digest
SHA256 1e9da41f0ad5ce7ecadd9d563413f3e05d6b26c01e2a6a44156803f2c3d85ace
MD5 c10b935ed402cffa42308d4490c07993
BLAKE2b-256 24935c43ab69cf13b4a683c238f03152c3c1031a14d72e6a44c9b5591997aa6b

See more details on using hashes here.

Provenance

The following attestation bundles were made for greatminds-2.2.0.tar.gz:

Publisher: publish.yml on veryviolet/greatminds

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file greatminds-2.2.0-py3-none-any.whl.

File metadata

  • Download URL: greatminds-2.2.0-py3-none-any.whl
  • Upload date:
  • Size: 382.8 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for greatminds-2.2.0-py3-none-any.whl
Algorithm Hash digest
SHA256 1ade24b104c3feac1faf7f919c2f4f6b8714043284f483c4bbf449d81cde3ac1
MD5 73d8921de2ffe769b237cb2664ced78a
BLAKE2b-256 57f38f2584ac7a91b8323c233ec7b4bc685a7d035a0b83c58e55ecabd1fc7b36

See more details on using hashes here.

Provenance

The following attestation bundles were made for greatminds-2.2.0-py3-none-any.whl:

Publisher: publish.yml on veryviolet/greatminds

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page