Skip to content

Smart Prompt v0.2.0-beta.1

Latest

Choose a tag to compare

@mumu-github mumu-github released this 07 Jun 13:38
· 0 commits to main since this release

Smart Prompt v0.2.0-beta.1

Release date: 2026-06-07

What Ships

  • Installable Windows beta package for Smart Prompt desktop shell.
  • Native local-service-sidecar.exe bundled with the installer, replacing the earlier Node.exe plus JS resources sidecar package.
  • Browser extension Prompt Card with manual idea, continue, and polish modes, Retry, Undo, Save, Insert verification, source badge, and redacted evidence.
  • Local service auth hardening, provider key migration, prompt/skill library APIs, metrics, backup/restore, diagnostics export, and clear-all-local-data flow.
  • Pilot loop covering ChatGPT, Claude, Gemini, desktop first-run, and adapter drift watch for Replit/Bolt/v0.

Artifacts

  • MSI: apps/desktop-shell/src-tauri/target/release/bundle/msi/Smart Prompt_0.2.0_x64_en-US.msi
  • NSIS setup: apps/desktop-shell/src-tauri/target/release/bundle/nsis/Smart Prompt_0.2.0_x64-setup.exe
  • Checksums: research/v5-beta-checksums.sha256

Verification

  • npm test in apps/local-service
  • npm test in apps/desktop-shell
  • npm test in prototypes/browser-extension
  • cargo check in apps/local-service-sidecar
  • cargo check in apps/desktop-shell/src-tauri
  • npm run build in apps/desktop-shell
  • powershell -NoProfile -ExecutionPolicy Bypass -File scripts\check-v4-installer-smoke.ps1
  • Native sidecar HTTP smoke: /health, /auth/bootstrap, /settings, /diagnostics/export, /data/all

Pilot Metrics

  • Insert success rate: 3/3 strict live-site Insert targets from V3 formal evidence.
  • Display baseline: 8/8 live-site display targets from V3 formal evidence.
  • Runtime Prompt Card baseline: Retry, Insert, Undo, online Save, and offline Save covered by extension runtime demo.
  • Desktop baseline: provider key save, provider test, skill import/delete, prompt save/delete, restart, diagnostics export, and clear local data covered by desktop interaction tests.

Known Boundaries

  • docs/prd.md has a user-unconfirmed local edit and is intentionally excluded from this release.
  • Claude validation depends on a logged-in persistent Chrome profile.
  • Replit formal evidence uses /agent4; /ai and root marketing textareas are rejected for formal composer evidence.