subagent-driven-development

An agent skill by obra, from obra/superpowers. Tags: api, developer-tools, mcp.

What it does

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

Install

With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:

npx skills add obra/superpowers --skill subagent-driven-development

Or copy the skill folder into Claude Code's skills directory by hand (~/.claude/skills for every project, or .claude/skills inside one):

git clone --depth 1 https://github.com/obra/superpowers
cp -r superpowers/skills/subagent-driven-development ~/.claude/skills/subagent-driven-development

Safety box score

Grade C, moderate risk: 35 of 100 risk points, rated 2026-10-20. The grade rates what the skill and its scripts can reach on the machine of whoever installs it, across eight categories from shell execution to secrets access. How the score works.

This skill orchestrates a multi-task implementation workflow by launching subagents, writing per-plan ledger/report artifacts under `.superpowers/sdd/`, and cleaning up the plan workspace at the end. The main risks are broad local execution plus autonomous progression through tasks without human check-ins, with recursive workspace deletion as the sharpest filesystem action.

Category levels, from 0 (none) to 3 (high)
Category	Level	Why
Code & shell execution	2 (moderate)	It shells out to local helpers like `bash scripts/sdd-workspace PLAN_FILE`, `bash scripts/task-brief PLAN_FILE N`, and `bash scripts/review-package PLAN_FILE BASE HEAD`, and the dispatched subagents are expected to run tests and commits.
Network access	0 (none)	No external URL fetches, API calls, or data exfiltration paths are described in the skill files.
Filesystem reach	2 (moderate)	It creates and uses plan-specific artifacts under `.superpowers/sdd/` and ends with `rm -rf <workspace>`; the setup also warns that `git clean -fdx` will destroy the scratch workspace.
Secrets & credentials	0 (none)	Nothing here asks for passwords, tokens, `.env` files, keychains, browser sessions, or other credential stores.
Privilege & persistence	0 (none)	There is no sudo, global install, service registration, or shell/OS settings change in the skill text.
Instruction hijack surface	2 (moderate)	It explicitly says `Do not pause to check in with your human partner between tasks` and tells the agent to `decide` ambiguities, which pressures it to skip normal confirmation.
Irreversible actions	2 (moderate)	The workflow is designed to execute an entire plan, dispatch subagents, review, and finish without stopping for user approval except for a few named cases.
Transparency	1 (low)	Core behavior is delegated to helper scripts (`scripts/sdd-workspace`, `scripts/task-brief`, `scripts/review-package`) that are referenced but not visible in the markdown evidence.

Before installing

Read `scripts/sdd-workspace`, `scripts/task-brief`, and `scripts/review-package` before trusting the execution surface.
Verify the workspace path really stays under `.superpowers/sdd/` and cannot resolve outside it during cleanup.
Try it first in a throwaway clone or scratch branch, because it is designed to run continuously and can delete the plan workspace.

Rated 2026-10-20 by an automated review of 4 files from the skill's repository.

Source

Repository: obra/superpowers (all skills from this repository)
Path: skills/subagent-driven-development/SKILL.md
Branch: main
Updated: 2026-09-19

diagnosing-superpowers — Use when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results.
dispatching-parallel-agents — Use when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies
atlassian-mcp — Integrates with Atlassian products to manage project tracking and documentation via MCP protocol.
claude-api — Reference for the Claude API / Anthropic SDK — model ids, pricing, params, streaming, tool use, MCP, agents, caching, token counting, model migration.
figma — Import Figma content into a HyperFrames composition — rendered assets, brand tokens, components.
hyperframes-registry — Search, install, and wire registry blocks and components into HyperFrames compositions.