bilingual_book_maker: Skill for Claude Code

bbm-plan: translate a book with bilingual_book_maker

This file is the router. Do the intake below, then load one flow file for the book in hand and follow it. The other flow files stay unread.

Repo: the repository this skill ships in (make_book.py at its root). Run every command from the repo root. The user documentation is the MkDocs site in docs/en/ of this checkout; the flow files point at the page to read for anything they do not spell out. python make_book.py --help is the authority on every flag, and docs/en/cmd.md lists them all in one place.

Your stance

You are the trained CLI operator. You pick every flag from the flow file's recommendations and state each choice with a one-line reason; the recommended defaults are what you pass unless the user asks for something else or the book argues otherwise. The user hands over a book, credentials and (if they have one) a prompt file, approves the plan and the cost, and gets a bilingual book back. They are never asked to experiment with flags, halt semantics or resume mechanics.

The EPUB flow uses --plan-classify agent. The plan arrives with its uncertain signatures set to "action": null, and the translate run refuses to start while any null remains, so you, the coding agent, own the classification against the real samples. Do it in the main agent with full session context; never delegate plan editing to a subagent or a small/fast model. The one exception is the user naming a classifier themselves (Jev, or a model): then follow "A classifier the user names" in references/epub-plan-mode.md, and still read its skips.

All state lives on disk (bbm_providers.json, .env, the flow's own working files, the resume cache, run.log), so any step can be redone after a crash or in a new session.

Intake

1. Route and credentials: always read references/route-setup.md before the first command. It probes what the user already has, asks one question (route, model, prompt file), binds the ROUTE and CONTEXT arrays every flow uses, and probes the endpoint for a sub-cent before anything is paid. Keys never go on the command line: the handoff block reprints the whole command, and argparse echoes it back on a mistyped flag. If a key ever reaches the terminal, say so at once and tell the user to rotate it.
2. Book path and target language (--language, e.g. zh-hans, ja, Simplified Chinese). For a small language the tables may not know, pass both halves: --language "ain:Ainu" (tag before the colon, name after; the tag list is docs/en/languages.md). --source_lang states the source language for a book whose short lines or names could be misdetected.
3. Their prompt file, asked in the route question. If they hand one over, lint it before the first paid run: references/prompt-files.md. If they say no, offer one sentence of what a style instruction would buy them and move on. Independently, look for an existing template in the book's directory and the repo root (prompt.json, prompt.txt, prompt*.md, prompt_template*). A candidate must carry a diff: in a git repo only untracked or modified-against-HEAD files count; cleanly tracked prompt* files are the repo's shipped examples, not the user's voice. Found one? Ask before doing anything with it; never adopt or ignore it silently.
4. Anything they want changed from the recommended defaults: a visual style for the translation, specific chapters or pages only.