Superpowers: How I'm using coding agents in October 2025 (blog.fsck.com)

Anthropic’s new plugin system for Claude Code set off the release of “Superpowers,” a plugin/skill framework the author built to turn Claude into a disciplined coding agent. Installable on Claude Code ~2.0.13 with two commands (plugin marketplace add obra/superpowers-marketplace; plugin install superpowers@superpowers-marketplace), Superpowers injects a session-start hook that teaches Claude to discover and use SKILL.md files, default to a brainstorm→plan→implement workflow, create git worktrees for parallel tasks, practice RED/GREEN TDD, dispatch subagents for implementation, run automated code reviews, and optionally open PRs or merge changes at the end. The kicker is skills: modular markdown-described capabilities that an agent can read, test, extend, and even generate for itself (the author taught Claude “how to create skills”). Superpowers includes pressure-testing of skills via realistic scenarios, checks for comprehensibility with subagents, and memory tooling that archives conversation transcripts, summarizes them with “Claude Haiku,” and indexes them in SQLite for retrieval. Technical implications are big: reproducible agent workflows, self-improvement by writing machine‑readable skills, built‑in code hygiene (TDD + reviews), and concerns around IP and subtle persuasion effects—research shows LLMs respond to influence principles, and Superpowers’ prompts intentionally use authority/commitment patterns to increase reliability. Sharing and integrated memory remain works-in-progress.