BloomDesktop: Skill for Claude Code

Voice a Bloom book with ElevenLabs

voiceBook.mjs records each text box of one language as a single ElevenLabs clip and writes the markup Bloom's spreadsheet import writes for audio recorded "By Whole Text Box" and then split (SpreadsheetImporter.AddAudioAsync). ElevenLabs' character timestamps supply the sentence boundaries, so Bloom highlights each sentence as it is read. Sentences come from Bloom's own splitter and checksums from Bloom's own getChecksum, both bundled from src/BloomBrowserUI at run time, so the Talking Book tool sees the recordings as current. Paths below are repo-root-relative; commands are bash.

1. Get the book folder

A book already in a collection: use its folder. A book on bloomlibrary.org: download it into a collection of its own, as Bloom's "Download into Bloom" would:

node .claude/skills/voice-book/downloadBook.mjs "https://bloomlibrary.org/language:swh/book/HRahJAASYF" "D:/some-folder"
# prints { collectionFile, bookFolder }

Done when you have a book folder holding the book's .htm.

2. Make sure Bloom is not holding the book

Bloom re-saves the book it has selected from its in-memory copy and silently overwrites outside edits. Close that Bloom, or select a different book in it, before running the script. Opening a book can also rename its folder to match the title (an old Bloom Library book came back named after its Swahili title), so list the collection folder again after Bloom has had the book.

3. Choose the voice and model

Ask the user which voice to use for this language, unless they already said. Voice ids from the account:

$h=@{'xi-api-key'=[Environment]::GetEnvironmentVariable('ELEVENLABS_KEY','User')}
(Invoke-RestMethod -Headers $h "https://api.elevenlabs.io/v2/voices?page_size=100").voices | % { "$($_.voice_id) | $($_.name) | $($_.labels.language) $($_.labels.accent)" }

The key lives in the Windows User environment as ELEVENLABS_KEY; a shell started before it was set does not have it, so read the User scope as above. The key needs the user_read permission, because the script reads the account's credit counter before and after voicing; without it the script stops before spending anything. Premade voices work on every account; library voices only on accounts that added them.