awman CLI
A Rust CLI and TUI: agents in containers, parallel tabs, TOML workflows from issue to PR and a scheduled agent squad
- 73 ★ on the repository
- Last commit
Broad access
Why this level
- Agents run unattended and can commit, push and open pull requests, and --yolo turns off their permission checks
- Steps get tokens and keys through env(...) overlays, so grant the minimum access
Install
- Works in
Manual install
curl -s https://prettysmart.dev/install/awman.sh | shThe installer detects the platform and puts awman on PATH.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Agents run unattended and can commit, push and open pull requests, and --yolo turns off their permission checks
- Steps get tokens and keys through env(...) overlays, so grant the minimum access
This is third-party code. Review the repository files before installing.
What it does
awman (Agent Workflow Manager) guides development with agents from an issue to a merged pull request. The first stage isolates an agent in a container and a git worktree. The second runs several agents in parallel in TUI tabs: Ctrl+T opens a tab with its own directory and container, and a tab turns yellow if an agent goes quiet for 30 seconds. The third stage turns a team process into repeatable workflows. A TOML or YAML file in the repository splits work into phases like plan, implement, review, docs, steps declare dependencies through `depends_on` and independent steps run in parallel. Each step names its agent and model, so Codex implements and Claude reviews, and declares overlays such as `skill(review)` or `env(API_KEY)`: only what is requested gets host access. Setup and teardown phases create a branch, run tests, commit, push, open a PR and wait for green CI, and an `on_failure` block launches an agent to fix problems. Run it with `awman exec workflow ./aspec/workflows/implement-pr.toml --work-item 0027`. The fourth stage is a squad of agents: `awman squad add` defines a task like "when a new issue is opened, analyze it and post a plan", and on an interval a squad agent decides whether the condition is met and then runs a workflow unattended. `awman api start` exposes HTTP so heavy workflows run on a build server. Supported agents are `claude`, `codex`, `opencode`, `gemini`, `antigravity`, `copilot`, `cline` and custom ones.
Who it is for. Teams and solo developers who run Claude Code, Codex and other agents on issues and want container isolation and a repeatable process up to the PR.
Good fit when
- Agents need isolation: each works in its own container and worktree and leaves the main copy alone
- A task goes through plan, implement, review and docs phases and the result needs checking between phases
- In one process Codex writes code and Claude reviews it
- New issues should be triaged automatically on a schedule with a plan posted as a comment
- Heavy runs should move to a build server and start remotely
Not a fit when
- You need one conversation with one agent and no infrastructure: Claude Code alone is enough
- There is no Docker or Podman: agents run in containers
- You need a yolo run without isolation: `--yolo` is designed for an isolated worktree
Example request
Run the implement-pr workflow for item 0027: Codex implements, Claude reviews, then tests, commit and a pull request.Limitations
Agents need a container environment, and the repository holds `.awman/Dockerfile.claude`, `Dockerfile.codex` and `Dockerfile.opencode`. The macOS build for Apple silicon includes an experimental builtin microVM runtime and is ad-hoc signed without notarization, so when downloading through a browser you must verify the SHA-256 and clear quarantine with `xattr -d com.apple.quarantine`. The installer and `mise` do not apply quarantine.
The `--yolo` flag turns on an agent mode with no permission checks and in `exec workflow` it automatically runs in a separate git worktree. The `--auto` flag approves file edits but still asks about shell commands. The `awman api start` server listens on loopback only, so access goes through an SSH tunnel. The license is Apache 2.0 and the last commit is from November 5, 2026.
How to disable. Remove the `awman` binary from PATH (for mise run `mise uninstall` for the package, for a source build delete the installed file). Stop the squad and the API server, then remove the `.awman/` directory from projects and `~/.awman/` from your home folder.
FAQ
How do I install awman?
Run `curl -s https://prettysmart.dev/install/awman.sh | sh`. There is also `mise use -g github:prettysmartdev/awman`, binaries on GitHub Releases or a source build with `sudo make install` on Rust 1.94 or newer.
How do I start in a project?
Run `awman init` once, then `awman` opens the TUI and the `chat` command starts an agent session. `ready --refresh` updates Dockerfile.dev so the project tools land in the image.
Which agents are supported?
Claude, Codex, OpenCode, Gemini, Antigravity, Copilot, Cline and custom agents. crush and maki ship as examples, and a workflow step picks its own agent and model.
What does yolo mode do?
The `--yolo` flag turns on the agent builtin no-permission-checks mode and keeps the workflow moving without you. In `exec workflow` it automatically runs in an isolated git worktree, and the final diff can be merged or discarded as a unit.
How does the squad work?
`awman squad start --background` starts the squad and `awman squad add` defines a task with an interval. On each interval an agent decides whether the condition is met and runs a workflow unattended.
Discussion
No questions or reviews yet.
Related
CLI toolsA server-side web job runner with an MCP endpoint: Claude Code runs jobs and workflows, reads their results and answers notes
A background process manager for agents: the agent starts a dev server and reads its logs, status and ports later, plus a TUI
A Claude Code sample: a lead agent, pools of coders, DevOps and reviewers, a spec and four test tiers
A background daemon loops over GitHub or Forgejo issues, drafts specs, reviews and fixes PRs until they are ready for human merge.
A git worktree CLI built for running Claude Code and Codex in parallel: the wt command, a plugin, and automation hooks
Agent of Empires
Agent of Empires (AoE)
A session manager for coding agents: runs Claude Code, OpenCode, Codex and other CLIs in parallel, isolated tmux sessions with a TUI and web dashboard
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| awman (this card) | no | 73 | ||
| container-use | yes | not required | 4,089 | gives an agent an isolated container environment, while this one adds phased workflows, TUI tabs and an agent squad up to a PR |
| Agent of Empires | no | 4,226 | manages several agent sessions, while this one builds repeatable workflows with access overlays and CI waiting |
README in short
The description builds the work in four stages: agent isolation, parallel sessions in the TUI, repeatable workflows and a squad for routine work. It then covers installation, quick start, details on workflows, yolo mode, the remote API and the skills library.