Skip to content

TeachReplay v0.2.0

Latest
Latest

Choose a tag to compare

View all tags
@Ottohere-Mourn Ottohere-Mourn released this 22 Aug 06:28
· 7 commits to main since this release
v0.2.0
36d0229

TeachReplay v0.2.0 — the standalone release

TeachReplay is now a harness-agnostic Teach-by-Demonstration engine: record a GUI/CLI
workflow once, compile it into a reusable, parameterized skill, and replay it with different
inputs — ending in an explicit success/failure verdict on real computers.

What changed since v0.1

  • Standalone monorepo — @teachreplay/core (trajectory schema, recorder, parameterized
    skill compiler, parameter substitution, replay engine, verifier, file stores,
    createTeachRuntime), @teachreplay/remote (generic SSH Linux computer backend),
    @teachreplay/mock (demo computer), @teachreplay/adapter-dsh — with zero
    OpenMausBot imports    , enforced by a test.
  • OpenMausBot is now an integration — the v0.1 application lives on as the thin
    reference adapter in TeachReplay-OpenMausBot.
  • DeepSeek Harness adapter verified — builds, registers teach_start/stop/compile/ replay/shell tools, and runs Teach → Compile → Replay through real DSH tool dispatch
    inside the official workspace (release 0.1.1-rc.2).
  • Standalone demo — pnpm demo: demonstrate → record → compile → change parameters →
    replay → verify, with no harness and no credentials.

Real evaluation (unchanged behavior, now on the standalone core)

8 tasks / 14 replays on a real remote Linux computer: 14/14 verdicts correct, 12/12
replay successes, 6/6 changed-parameter replays, 2/2 explicit failure detections, 0
ground-truth mismatches — identical to the v0.1 integration results.

Notes

  • Attribution: Apache-2.0; derived from TeachReplay v0.1 inside OpenMausBot (NOTICE).
  • npm packages are not published yet — install from source (pnpm install && pnpm build).
Assets 2
Loading