subagent-driven-development Skill for Claude: What It Does and How to Install It

Makes Claude run a written plan by giving each task to a fresh implementer subagent, then reviewing every task and the whole branch.

Installs: 244.8K; on skills.sh
Stars: 310.6K; on GitHub
Files: 13; besides SKILL.md
Updated: Oct 2026; last repo commit

subagent-driven-development makes Claude coordinate instead of code. For each task it briefs a fresh implementer subagent, builds a review package, and sends a task reviewer to check spec compliance and code quality. A fix loop repeats until both pass. A broad final review of the whole branch comes last. Subagents never inherit your session history, which keeps Claude's own context free.

Like executing-plans it keeps a ledger, does not pause between tasks, and stops only for destructive actions, security-sensitive actions, shared side effects such as a push or merge, or a broken plan. It costs a fresh context per task and per review. It bundles three shell scripts and several prompt and reference files, and it refers to other superpowers skills, so install the whole superpowers set.

Good for

Developers with a written plan of mostly independent tasks
Teams that want a review gate on every task
Developers who want to choose which model each subagent role uses

Install subagent-driven-development

Claude Code

Copy this prompt.

Install the subagent-driven-development skill in Claude Code. Run this command:

/plugin install superpowers@claude-plugins-official

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

This installs the whole superpowers plugin. subagent-driven-development calls sibling skills and scripts from the same plugin, so installing it alone leaves gaps.

From the project README. Source.

What's in the skill

When it loads

Use when executing implementation plans with independent tasks in the current session

Sections in SKILL.md

When to Use
The Process
Setup
Model Selection
The Task Loop
1. Dispatch the implementer
2. Handle the report
3. Review the task
4. The fix loop
5. Complete the task
Final Review
Common Rationalizations

Key facts

Author: obra
Repo: obra/superpowers (skills/subagent-driven-development)
Installs on skills.sh: 244,830 (Nov 2026)
GitHub stars: 310,635 (Nov 2026)
License: MIT
Last repo commit: Oct 29, 2026
Files in the skill folder: 14 including SKILL.md
skills.sh audits: Gen Agent Trust Hub: pass, Socket: pass, Snyk: pass
Scripts: sdd-workspace, task-brief and review-package

What it includes

Uses an isolated worktree and a plan workspace with a ledger (scripts/sdd-workspace)
Runs code · Setup
Extracts the task with scripts/task-brief and sends it with implementer-prompt.md
Runs code · Dispatch the implementer
Reads the implementer's status and follows references for anything other than DONE
Handle the report
Builds a review package with scripts/review-package for a spec and quality reviewer
Runs code · Task review
Repeats fixes and re-reviews until the reviewer approves
Runs code · Fix loop
A broad review of the whole branch before finishing
Read · Final review
Hands off to the finishing-a-development-branch skill
Finish

From the official docs

Permissions

Medium risk

Claude runs bundled scripts, dispatches subagents and makes commits across the whole plan without pausing between tasks. Run it in a worktree on a branch and review the commits before you merge.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

It keeps going without checking in between tasks
It stops only for destructive or irreversible actions, security-sensitive actions, shared side effects like a push or merge, and broken plans.
It runs three shell scripts
sdd-workspace, task-brief and review-package create ledger and review files in the repo. Read them if you want to know exactly what they write.
Subagents write and commit code
Implementer subagents implement, test and commit. Review the commits afterwards.
It deletes the plan's workspace
When the final review is clean, the process in SKILL.md deletes this plan's workspace directory (the git-ignored ledger, briefs and review packages).

Questions

How do I install subagent-driven-development?

Use the install box above. It installs the whole superpowers set, because this skill uses sibling skills such as using-git-worktrees and finishing-a-development-branch.

subagent-driven-development or executing-plans?

Use this one when you want a fresh implementer and a review after every task. Use executing-plans when you chose inline execution or have no subagent tool. SKILL.md says both share the same plan workspace and ledger.

What does it cost?

A fresh context per task and per review, according to SKILL.md. Inline execution costs one context plus one final reviewer.

Does it need a plan?

Yes. It executes an implementation plan whose tasks are mostly independent. If tasks are tightly coupled, SKILL.md points to manual execution or brainstorming first.

Is it safe?

It runs shell scripts, subagents and git commits in your repo without pausing between tasks. Run it in a worktree on a branch, and review the commits and ledger before you merge or push.

Information

updated Nov 2, 2026

Repository: obra/superpowers
Folder: skills/subagent-driven-development

obra/superpowers skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 2

obra/superpowers README · github.com
subagent-driven-development SKILL.md · github.com