November 20, 2026SkillsCodingOpen Source

Superpowers v7: The Skill Pack That Learned to Ask Why

obra/superpowers is back on GitHub Trending, and hard. It added 2,029 stars on Thursday's daily board, up from 598 a day earlier, and now sits at about 322,000 stars under MIT. Repo: https://github.com/obra/superpowers

Nothing shipped in our window to explain the jump. The last release, v7.0.2, landed on November 3 and the last push was the same day. So the spike is word of mouth, a big link somewhere we could not find, or people finally noticing what changed since we last covered it at v6.4.1 in September. The stars are the news hook. The real story is the v7 line.

v7.0.0 (October 10) threw out the brainstorming skill and rebuilt it. Say what you want and why, it plays back what it heard so you can correct it, keeps asking until the picture is clear, then sizes the process to the job, from just doing it to writing a full design. Users pushed back fast: the new default questions felt formulaic. v7.0.1 (October 21) made it read the project and any spec first and stop asking what the files already answer. v7.0.2 (November 3) goes further: it asks only when the answer would change what gets built, asks why first if you never said, and when there is a real choice it lays out options with a recommendation instead of a bare question. It also trimmed three long skills so they survive Claude Code compaction, and fixed bootstrap on six harnesses. Release notes: https://github.com/obra/superpowers/releases

This is the most useful thing a skill pack has published all year, and it is not code. It is a changelog about how an agent should interrupt a human. Too many questions and people skip the skill. Too few and the agent builds the wrong thing confidently. Superpowers is tuning that dial in public, release by release, with user complaints as the eval. Everyone shipping an agent has the same dial and most are not even measuring it.
← Previous
Intent Contracts and Idempotency: Controls Outside the Model, With the Bill Attached
Next β†’
Three Papers, One Bill: Where Agent Tokens Actually Go
← Back to all articles

Comments

Loading...
>_