i18nlisted

# i18n — documentation translation Translation runs as a pipeline: a script plans the work and hands out chunks, **subagents do the actual translating**, and a script reassembles and verifies. You orchestrate; you do not translate whole documents yourself in the main thread. Everything is this skill's own code. `run.sh` runs each script through `uv`, which supplies its two dependencies (`markdown-it-py` for CommonMark-accurate fence detection, `pyyaml` for YAML resource files) per-run. Always invoke the scripts through `run.sh`. ## Hard rules - **Never** translate: fenced or inline code, URLs, placeholder tokens (`{count}`, `%s`, `${VAR}`), resource keys, frontmatter keys, `@@CODE_BLOCK_n@@`-style tokens. - **Never** hand-write a translated file. Everything goes through `run.sh apply`. - **Never** accept output that `run.sh verify` rejects. Fix it or report it. - **Never** overwrite a translation a human edited. `plan` reports these as conflicts; ask the user before passing `--force`. - Decline requests to translate source-code comments or in-code strings; say why and offer to translate the docs instead. - No external translation API, ever. Subagents are the translator. ## Quick start ```bash S=~/.claude/skills/i18n/scripts # Codex: ~/.codex/skills/i18n/scripts $S/run.sh plan --root . --lang zh-CN --paths 'README.md' 'docs/**/*.md' # ... fan out one subagent per file in <state-dir>/work/<run>/tasks/ ... $S/run.sh apply --root . --run <run_id> $S/run.sh