Dev & Engineering multi-agent-cliterminal-workspacetmuxcodex-cligemini-cliandroid-remote-control

CCB Multi-Agent Terminal Workspace

Coordinate multiple AI coding CLIs in visible terminal panes you can directly take over.

FollowAgents review · FARS-2.1
Use with care
64/ 100 5-point scale 3.2 / 5
1 2 3 4 5 6
Per-dimension scores and reasoning
1Trust18 / 29 · 3.1/5

Evidence shows: Config UI binds to loopback, tokens provided via env or file, token value never printed; mobile gateway binds loopback by default, LAN binding requires specific private interface, remote access uses Tailscale Serve not Funnel; does not store Tailscale passwords or OAuth tokens; update flow has transactional rollback and recovery backup path; diagnosis requires user authorization before submitting GitHub issues. Deductions: dependency security only partial evidence (package.json has no dependency lock, no dependency audit), external effects limited but not fully verified, source attribution only via README and contact, publisher identity not verified.

2Reliability8 / 14 · 2.9/5

Evidence shows: README and code mention fail-closed and recovery mechanisms, e.g., v8.5.1 Claude incomplete replies fail closed, v8.5.2 recovery has 90-second probation and backoff; test scripts cover communication matrix and stress tests. Deductions: dependency availability not fully verified as no dependency manifest or lock file, and external CLIs like tmux may be unavailable; failure messages present in code but not fully assessed.

3Adaptability12 / 18 · 3.3/5

Evidence shows: README clearly targets developers, supports multiple scenarios (multi-agent collaboration, mobile control, rich media terminal); capability boundaries defined via config and role packs; trigger precision via commands and config; environment fit supports Linux, macOS, WSL, and handles paths and permissions. Deductions: capability boundaries and trigger precision described in docs but no detailed examples or boundary tests; environment fit mentioned but not all possible environments covered.

4Convention12 / 18 · 3.3/5

Evidence shows: README well-structured with quick start, config, mobile app, roles, release notes; install notes detailed including npm, GitHub release, and source install; naming stable (ccb, ask); examples and FAQ via README and docs; known limitations in release notes (e.g., v8.5.5 withdrawn); license AGPL-3.0; versioning and changelog in README; maintenance responsibility via contact. Deductions: information architecture good but some docs may be incomplete; install notes detailed but no troubleshooting; naming stability not fully verified; examples and FAQ may be insufficient; known limitations not exhaustive; license metadata NOASSERTION but actual AGPL-3.0; versioning and changelog in README but no separate file; maintenance responsibility clear but not verified.

5Effectiveness9 / 13 · 3.5/5

Evidence shows: Output usability via terminal UI and mobile app, supports file transfer and remote terminal; marginal value via multi-agent collaboration and cross-provider integration; cost-benefit via free open-source and lightweight design. Deductions: output usability not actually tested, marginal value may be limited by external CLI dependencies, cost-benefit not quantified.

6Verifiability5 / 8 · 3.1/5

Evidence shows: Claims in README partially supported by code and tests, e.g., release notes features correspond to code; cross-source verification via test scripts and CI workflows; fact-inference separation in docs, e.g., distinguishing features and known limitations. Deductions: claim traceability not fully verified as no specific code references; cross-source verification only via internal tests, not independent; fact-inference separation in docs but not explicitly labeled.

Evidence confidence: Low Reviewed Aug 09, 2026 Reviewed revision a8d119ce9472
The upstream repository has new commits since this review. The score still applies to the reviewed revision shown and may not cover the latest changes.
Before you use it
  • Dependency security: package.json lacks dependency lock or audit, recommend checking for vulnerabilities.
  • External effects: mobile gateway and remote access have restrictions but ensure correct configuration to avoid unintended exposure.
  • Publisher identity not verified, verify source before installation.
Review evidence [1][2][3][4][5][6][7][8]
See the full review method →

What does this agent do, and when should you use it?

CCB is a developer-focused terminal UI and background daemon operated through the `ccb` command. It coordinates Codex, Claude, Gemini, and other supported CLI providers within one project while keeping each agent in a visible native terminal pane. Project topology is stored in `.ccb/ccb.config`, and durable shared coordination can live in `.ccb/ccb_memory.md`; the configuration panel manages windows, splits, providers, models, and workspaces. Its daemon is intended to preserve project state after the foreground interface closes. The repository also includes the Flutter-based Android CCB Mobile app, which connects through an authenticated gateway for project viewing, pane-native input, terminal access, and file transfer.

Running ccb from a project directory starts a workspace. For a blank project, CCB opens one demo agent and selects the first supported CLI found locally. Users can open ccb config ui or use the sidebar settings panel to save .ccb/ccb.config; its [windows] section can define layouts such as worker1:codex(worktree), worker2:claude(worktree), and ccb config validate checks the file before launch. Work can be entered directly in a pane or delegated with /ask reviewer review the latest parser changes and list blocking issues. Shared, long-lived handoff context belongs in .ccb/ccb_memory.md. ccb update updates CCB and can check or update supported provider CLIs, while ccb update mobile configures Android remote control and ccb update rich enables the Rich WezTerm launch mode.

  1. A development team using Codex, Claude, and Gemini CLIs that wants to inspect each agent’s terminal and hand work between them in one project.
  2. A maintainer who wants Codex coding in a worktree while Claude reviews changes and Gemini performs QA in a saved pane layout.
  3. An engineer on Linux, macOS, or WSL who needs project state to remain available after closing the foreground terminal UI.
  4. A developer who needs to inspect mounted projects, send text to a selected pane, or open a terminal from an Android phone.
  5. A user who wants to configure provider selection, models, thinking levels, pane splits, and workspaces through a local control panel.

What are this agent's strengths and limitations?

Pros
  • Coordinates multiple provider CLIs in one visible terminal layout, with direct pane takeover rather than only background orchestration.
  • Persists project topology in .ccb/ccb.config and offers a configuration UI with validation, reload dry-runs, and guarded hot reload.
  • Keeps project state in a background daemon when the foreground TUI is closed.
  • Includes an Android remote-control path with authenticated project discovery, pane input, terminal access, and file transfer, plus documented loopback, private-LAN, and Tailscale Serve boundaries.
Limitations
  • It depends on locally installed, working provider CLIs; the initial agent selection depends on which supported CLI is detected first.
  • The documented platform scope is Linux, macOS, and WSL; native Windows support is not described.
  • Advanced teams must maintain a project-level .ccb/ccb.config and understand its window and split syntax.
  • Mobile delivery is currently centered on an Android APK, and LAN or remote use requires private-interface or Tailscale Serve configuration.
  • The supplied repository metadata reports NOASSERTION for the license, so usable license terms are not established here.

How do you install or deploy this agent?

Install with npm: npm install -g @seemseam/ccb@latest. Then change to a working directory and run ccb. If startup says that the project anchor cannot be created automatically, run mkdir -p .ccb and retry. Install at least one supported provider CLI locally; CCB selects the first available one. As alternatives to npm, a matching GitHub Releases package can be unpacked and installed with ./install.sh install; source installation is documented for development or temporary fallback.

How do you use this agent?

From the project directory, run ccb; the first blank-project launch opens the default demo agent. To persist a multi-pane setup, run ccb config ui, or create .ccb/ccb.config, run ccb config validate, and start ccb. Type in an agent pane directly or dispatch to a named agent with /ask <agent> <task>, for example /ask reviewer review the latest parser changes and list blocking issues. Store team rules and handoff context in .ccb/ccb_memory.md. For Android control, run ccb update mobile; the gateway defaults to loopback, LAN mode accepts a specified private interface address, and remote access uses Tailscale Serve.

FAQ

Does CCB provide model accounts or credentials for Codex, Claude, or Gemini?
No. It coordinates supported provider CLIs available on the local machine; installation and authentication for those CLIs come from the user environment.
Does closing the CCB interface stop the workspace?
The README states that its background daemon keeps project state alive when the foreground UI is closed.
Where should a team keep persistent coordination rules?
Use the project-wide .ccb/ccb_memory.md file for rules, constraints, durable context, and handoff conventions.
Can the mobile gateway be exposed directly to the public internet?
It binds to loopback by default. LAN mode requires one specified private interface address, while remote access uses Tailscale Serve rather than Tailscale Funnel.

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related agents