FIND-20260323-033 · 2026-03-23 · Innovation Veille
obra/superpowers — Agentic Skills Framework for Structured Multi-Agent Development
adhoc
HIGH
superpowers is a plugin-based skills framework for AI coding agents (Claude Code, Cursor, Codex, Gemini CLI) that imposes a structured software development methodology: brainstorming → planning → subagent-driven implementation → TDD → review → branch completion. Each skill is a SKILL.md document in ~/.claude/skills/ that gets injected via a SessionStart hook. The repo ships skills for dispatching parallel agents, subagent-driven development, systematic debugging, test-driven development, git worktrees, plan writing, and code review handoff. With 107k stars and active maintenance (last commit 2026-03-19), it is the most-starred agentic methodology framework on GitHub as of March 2026.
Source
https://github.com/obra/superpowers
ODS Impact
ODS ADLC pipeline already implements an identical architecture (dispatcher-v3.sh + Claude tmux subagents + SKILL.md files in ~/.claude/skills/). superpowers provides a battle-tested, community-validated reference implementation of the same patterns. Specific adoptable patterns: (1) the two-stage review loop per task (spec compliance then code quality) from subagent-driven-development/SKILL.md — ADLC currently runs ba/architect/security/devops as separate gates, not per-task inline loops; (2) the model-selection strategy matching cheapest model to mechanical tasks and most capable to architecture/review; (3) the four status codes DONE / DONE_WITH_CONCERNS / NEEDS_CONTEXT / BLOCKED for subagent responses — ADLC only uses PASS/FAIL today; (4) the session-start hook injecting skill context automatically on every Claude Code session startup — ODS skills are loaded manually today; (5) the dispatching-parallel-agents skill formalizing domain isolation criteria and conflict-free integration patterns. The git worktrees skill is also directly applicable for the ADLC parallel dev pattern to avoid branch conflicts.
Security Review
License: MIT | Maintenance: ACTIVE | Risk: LOW | Recommendation: SAFE_TO_USE
Tags
agents
claude-code
skills
subagents
methodology
tdd
adlc
orchestration
hooks