# Changelog

All notable changes to this project are documented in this file.

The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).

## [1.1.1] - 2026-09-04

### Fixed

- Portable Python resolution: the venv path is now derived from `$HOME`
  (`~/.dsh-sensevoice/venv`, matching `scripts/setup.ps1`) with a
  `DSHSENSEVOICE_PYTHON` env override, instead of hard-coded author paths.
- Dropped the ignored `cache_dir` argument; the model now always lives in the
  funasr default cache (`~/.cache/modelscope`).

### Changed

- README restructured for brevity (install/uninstall front and center,
  architecture diagram replaced with a one-line flow).
- Bundle row id aligned to the package name (`dshtools-sensevoice-input`) and
  `verify:package` now enforces the id == name convention for marketplace
  spec compliance; GitHub topics added for catalog discovery.

## [1.1.0] - 2026-09-04

### Changed

- Repository restructured into a standard TypeScript layout: TypeScript source
  (`src/`), tsdown dual-face build (ESM host + factory-form client bundle),
  vitest tests, CI/publish workflows, and the usual community files
  (CONTRIBUTING, SECURITY, issue templates, bilingual README).
- Built `lib/` output is committed to git so `dsh plugin add github:...`
  installs work without a local build step.

## [1.0.1] - 2026-09-04

### Fixed

- Host row now declares `inject: ['webServer', 'subprocess']` instead of
  non-blocking `ctx.get('webServer')` reads. The desktop mounts its webserver
  row after profile bundles, so apply() previously ran before the service
  existed and silently skipped route registration — every request fell through
  to the SPA static handler (`HTTP 405` on transcribe).

### Added

- README troubleshooting table, setup script (`scripts/setup.ps1`), repository
  metadata and GitHub install instructions.

## [1.0.0] - 2026-09-03

### Added

- First release: mic button in the composer tool row
  (`conversation.input.left`), recognition status notice
  (`conversation.input.overlay`), local SenseVoiceSmall (funasr) CPU inference
  behind `/sensevoice/*` HTTP routes, draft writing via
  `inputActions.setDraft()`, language/emotion tags, lazy model loading,
  subprocess self-healing.