subtask
skill
Gecti
Health Gecti
- License — License: MIT
- Description — Repository has a description
- Active repo — Last push 0 days ago
- Community trust — 327 GitHub stars
Code Gecti
- Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Gecti
- Permissions — No dangerous permissions requested
Purpose
This tool provides a CLI and a Claude Code "skill" designed to manage parallel AI subagents. It uses Git worktrees to isolate tasks, allowing a main AI agent to spawn, track, and interact with subagents working on different features or bug fixes simultaneously.
Security Assessment
Overall Risk: Medium. The code itself passes a light static scan with no dangerous patterns, hardcoded secrets, or dangerous permissions requested. However, the recommended installation method pipes a remote script directly into the shell (`curl ... | bash`). While common and optional (a safer Homebrew or Go install is available), executing unverified remote scripts is a potential supply chain risk. Furthermore, by design, the tool orchestrates and executes shell commands, spawns AI subagents, and heavily interacts with your local file system and Git repositories.
Quality Assessment
The project demonstrates strong fundamentals. It is licensed under the permissive MIT license and is actively maintained (last updated recently). It has earned a solid 327 GitHub stars, indicating a good level of community trust and interest. The documentation is clear and well-structured, featuring a transparent guide for usage and installation.
Verdict
Use with caution: The underlying Go code is clean and well-received, but you should avoid the `curl | bash` installation method in favor of Homebrew or the pre-compiled binary to ensure maximum safety.
This tool provides a CLI and a Claude Code "skill" designed to manage parallel AI subagents. It uses Git worktrees to isolate tasks, allowing a main AI agent to spawn, track, and interact with subagents working on different features or bug fixes simultaneously.
Security Assessment
Overall Risk: Medium. The code itself passes a light static scan with no dangerous patterns, hardcoded secrets, or dangerous permissions requested. However, the recommended installation method pipes a remote script directly into the shell (`curl ... | bash`). While common and optional (a safer Homebrew or Go install is available), executing unverified remote scripts is a potential supply chain risk. Furthermore, by design, the tool orchestrates and executes shell commands, spawns AI subagents, and heavily interacts with your local file system and Git repositories.
Quality Assessment
The project demonstrates strong fundamentals. It is licensed under the permissive MIT license and is actively maintained (last updated recently). It has earned a solid 327 GitHub stars, indicating a good level of community trust and interest. The documentation is clear and well-structured, featuring a transparent guide for usage and installation.
Verdict
Use with caution: The underlying Go code is clean and well-received, but you should avoid the `curl | bash` installation method in favor of Homebrew or the pre-compiled binary to ensure maximum safety.
Claude Skill to do your tasks with subagents in Git worktrees
README.md
Subtask ✨
Subtask gives Claude Code a Skill and CLI to create tasks, spawn subagents, track progress, review and request changes.
- Each task gets a Git worktree, so they can be done in parallel safely
- Claude can interrupt and talk with subagents!
- TUI shows progress, diffs & conversations
- Tasks are persisted in folders
- Codex subagents supported
- Ralph not needed
Spawning Tasks
You: fix the auth bug and add API metrics with Subtask
Claude:
├─► Skill(Subtask)
├─► Bash(subtask draft fix/auth-bug ...)
├─► Bash(subtask draft feat/api-metrics ...)
└─⏺ Tasks are in progress! I'll get notified when they're done.
Tracking & Merging
You: which tasks are open?
Claude:
├─► Bash(subtask list)
│ ┌──────────────────┬─────────┬────────────────────────┐
│ │ TASK │ STATUS │ TITLE │
│ ├──────────────────┼─────────┼────────────────────────┤
│ │ fix/auth-bug │ replied │ Fix auth token refresh │
│ │ feat/api-metrics │ working │ Add API metrics │
│ │ feat/ui-redesign │ draft │ Redesign dashboard │
│ └──────────────────┴─────────┴────────────────────────┘
│
└─⏺ ui-redesign hasn't started yet — want me to kick it off?
You: lets merge fix/auth-bug.
TUI
Run subtask in your terminal to see everything:
Setup
[!NOTE]
Subtask is in early development. Upcoming releases will simplify installation, solve known bugs, and improve Claude's proficiency.
Install the CLI
Mac/Linux
curl -fsSL https://subtask.dev/install.sh | bash
Windows (PowerShell)
irm https://subtask.dev/install.ps1 | iex
Other install methods…
Homebrew
brew install zippoxer/tap/subtask
Go
go install github.com/zippoxer/subtask/cmd/subtask@latest
Binary
Install the Skill
Tell Claude Code:
Setup Subtask with `subtask install --guide`.
Claude will install the Subtask skill at ~/.claude/skills, and ask you whether subagents should run Claude, Codex or OpenCode.
subtask install
# Tip: Uninstall later with `subtask uninstall`.
Install the Plugin (Optional)
In Claude Code:
/plugin marketplace add zippoxer/subtask
/plugin install subtask@subtask
This reminds Claude to use the Subtask skill when it invokes the CLI.
Use
Talk with Claude Code about what you want done, and then ask it to use Subtask.
Examples:
"fix the login bug with Subtask""lets do these 3 features with Subtask""plan and implement the new API endpoint with Subtask"
What happens next:
- Claude Code creates tasks and runs subagents to do them simultaneously.
- Claude gets notified when they're done, and reviews the code.
- Claude asks if you want to merge, or ask for changes.
Updating
subtask update --check
subtask update
Subtask is Built with Subtask
- I use Claude Code to lead the development (i talk, it creates tasks and tracks everything)
- I use Codex for subagents (just preference, Claude Code works too)
- ~60 tasks merged in the past week
- Proof
License
Yorumlar (0)
Yorum birakmak icin giris yap.
Yorum birakSonuc bulunamadi