Skillreaper hit Hacker News this morning as a Show HN from the thousandflowers GitHub account โ€” an open-source diagnostic aimed at AI agent setups, designed to surface skills that load into every session but never actually get invoked by the model. The premise is simple: if it's in your context window every single time and your agent has never once fired it, that skill is dead weight. Anyone who runs a serious agent deployment knows this smell. Skills accumulate like browser extensions โ€” you install one for a specific job months ago, it registers at startup, and it just keeps riding along session after session, consuming context budget and adding latency to every cold start while contributing exactly nothing. The HN listing is fresh off the press with minimal engagement so far, which tracks for a niche dev tool on day one. The repo URL points at github.com/thousandflowers/skillreaper โ€” but no release notes or README content were recoverable from our source pull, so exact detection mechanics (log analysis? manifest diffing? runtime instrumentation?) remain unconfirmed until the docs are readable. That uncertainty aside, the problem space is real and getting worse. Agent skill libraries have exploded over the past year as frameworks like OpenClaw made plugin ecosystems a first-class citizen โ€” and with that explosion came session bloat nobody was tracking. Skillreaper is attacking an inefficiency most operators haven't even quantified yet.

Key Takeaways

  • Skillreaper targets agent skills registered at startup that never fire in practice โ€” pure context-window overhead for real deployments.
  • The project is early-stage: a bare Show HN post with little traction, and the README wasn't parseable from our fetch. Treat it as v0.1 territory until docs confirm how detection actually works.

The Bottom Line

If you run agents with a growing skill library, this is exactly the kind of housekeeping tool worth watching โ€” dead skills are tax without representation in your token budget. Just wait for the README to load before wiring it into production.