Repository navigation
Review Voice Kit - #2584
Review Voice Kit#2584
Conversation
Pin swordfeng/paseo-voice-kit (plugin/) at 82023555b87b and remove submissions/2583.json. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
Voice Kit is approved, @swordfeng, and will be published with the next registry build. Nothing more is needed from you.
I pinned swordfeng/paseo-voice-kit (plugin/) at 82023555b87b (current main, since no reference was needed) as swordfeng/voice-kit, with categories Extras and Utils, and removed the intake file. Your OVERVIEW.md (plus the zh-CN one), description, translations and four screenshots are used as shipped.
Next-release advice (optional): the speech server in the v0.1.1 release was built from the tag (2e95cae), while this commit is 26 commits later (docs, workflow, a small server/ cleanup). Tagging the commit you release, and cutting the next server release from it, makes the binary and the source I can read line up exactly.
Review data
- Artifact: git
https://github.com/swordfeng/paseo-voice-kit.git@82023555b87b70587b3cca6b2f864393b43176e4, pathplugin. Fetched at the exact commit without submodules or hooks; nothing was installed, built or run. - Read:
index.server.ts, all ofserver/,shared/, plus a search ofclient/for network, exec and storage use. Build isnpm ci --omit=dev --ignore-scripts;npm-shrinkwrap.jsonv3, 349 entries, all from registry.npmjs.org with integrity; production deps are only@getpaseo/clientand its dependencies; the only install-script packages (esbuild, fsevents) are dev-only. - Runtime download: the plugin downloads the speech server archive from
github.com/swordfeng/paseo-voice-kit/releases/download/v0.1.1/only after the user approves it in a prompt, and checks a pinned SHA-256 before unpacking (server/binary.ts). I downloaded linux-x64 and confirmed73431f14…2f1bmatches the pin, and GitHub's published digests for all five archives matchshared/release.ts. The archive extractor (server/untar.ts) rejects absolute paths,..and non-file entries. Models download from huggingface.co and github.com with checksums inserver/catalog.json(13 entries). - Execution: spawns the downloaded (or user-chosen)
voicekit servebound to 127.0.0.1 with a per-run token;ps/powershellcalls only to find and stop its own leftover servers;paseo reloadafter writingproviders.local.modelsDir. - Credentials and config: reads the daemon
local-credentialonly to connect to the local daemon for the shared model library; editsconfig.jsonto point speech at the local server and parks the previous settings in a 0600 state file, restored when turned off (overview says so). Custom remote-server API keys are stored 0600 and never returned to the app. - Not inspected: the native speech-server binaries themselves. Their source is in the repo and CI builds them from a tag; I only checked the archive, checksum and the source diff between the release tag and the pinned commit (non-functional for the binary: catalog text, an unused persona path removed, workspace setting).
- Screenshot
voice-kit-speaking.pngviewed; it shows the plugin's settings page. Validation (node scripts/validate.tson main's validator): 557 records well-formed.
Closes #2583. Submitted by @swordfeng.
Ready for artifact review. The reviewer will resolve the submitted source and complete the registry listing.
Submission:
submissions/2583.json.Git references are optional. Use a supplied reference; otherwise resolve main during review. Git updates are submitted manually.