LearnHarness hits 1k stars: open-source agent harness that teaches from what you already know
burny_tech · x · 2026-10-25
LearnHarness is an open-source project (AGPL-3.0) that just crossed 1k stars — an "agent harness for learning" that uses AI agents to plan and teach lessons from your current understanding to a goal.
The methodology has three steps:
- Probe: the agent probes your understanding of everything the lesson depends on, building a detailed, continually updated picture of your mental models.
- Plan: it reasons out the minimal dependency tree from what you know to the goal.
- Teach: starting from what you already understand, it teaches one reasoning step at a time, aiming for each step — and the goal — to feel obviously true.
The premise: the best explanations and practice problems depend on what and how you already understand, so the agent needs a detailed map of that understanding. Ships with an Electron desktop app and a built-in learnharness skill.
More from coding & agent
- POA chain restarts from checkpoint 44003 after overload and duplicate-transaction outage — adamamcbride · 2026-10-25
- Dev calls out Morph for fake waitlist tactics: call first, then demand billing info — arthurcolle · 2026-10-25
- Why Formal Verification Is Exploding Now: Proofs Get Cheaper, Exploits Costlier — burny_tech · 2026-10-25
- Codex roadmap leak: agent message board, artifacts, code mode and more coming soon — umeshonai · 2026-10-25
- Claude reverse-engineered Ultima Online files, spun up a private server, and plays it with its owner — PestoPastaLover · 2026-10-25
- The age of humans reviewing code directly is largely over, argues @0xdoug — StefanoGogioso · 2026-10-25