Yuque Doc Push

Yuque Document Push

Manage documents in a Yuque knowledge base through the OpenAPI. Owns create, read, update, delete of documents and TOC inspection.

What this skill owns

• CRUD operations on documents within a single Yuque knowledge base.
• Automatic TOC insertion after document creation.
• Reading and displaying the knowledge base table of contents.

This skill does NOT own:

• Yuque repository or team management (create/delete repos, manage members).
• Cross-repo operations or batch migration.
• Yuque authentication setup (the user must provide a valid token).

Prerequisites

1. Python 3.8+ with requests and python-dotenv installed.
2. A .env file in the project root (created via the setup flow below).

First-time setup

If the user has not configured this skill yet (no .env file, or list returns a config error), follow this flow:

1. Ask for the knowledge base URL. Tell the user:

   Please paste your Yuque knowledge base URL, for example: https://xxx.yuque.com/group/book

   You can find it by opening your knowledge base in the browser and copying the address bar URL.

2. Ask for the API token. Tell the user:

   Please paste your Yuque API token.

   To create one: open Yuque -> click your avatar (top right) -> "Account Settings" -> "Tokens" -> "Create new token". Grant it read/write access to your target knowledge base.

   Token page: https://www.yuque.com/settings/tokens

3. Run the setup command. The script auto-parses the URL and validates the connection:

   python scripts/yuque_cli.py setup --url "" --token ""
   

   • On success: .env is written, connection is verified, .gitignore is checked.
   • On failure: the script prints the exact error (bad token, wrong URL, etc.). Fix and retry.
   • Use --force to overwrite an existing .env.

4. Confirm by running python scripts/yuque_cli.py list.

Security rules

• NEVER echo, print, or include the token in your text responses to the user. Pass it only as a CLI argument to the setup script.
• NEVER commit .env to git. The setup script checks .gitignore and warns if .env is not excluded.
• NEVER hardcode the token in any file other than .env.
• If the user pastes a token in chat, immediately use it in the setup command and do not repeat it back.

Default path

1. Validate prerequisites. Run python scripts/yuque_cli.py list to confirm token and repo are valid.
2. Identify the action. Determine whether the user wants to create, update, delete, or inspect.
3. Execute the action using scripts/yuque_cli.py with the matching subcommand.
4. Verify the result. Use list, get, or toc to confirm the operation succeeded.
5. Report the outcome: doc ID, title, slug, and any errors.

Command reference

Global flags available on all commands: --json (machine-readable output), --dry-run (preview without executing).

Dry-run mode always outputs JSON regardless of the --json flag. Use it to inspect the payload before committing.

Sync workflow

Use sync for "push only what actually changed" instead of blind update. The script maintains .yuque-sync.json (auto-added to .gitignore) tracking each doc's local content hash and remote latest_version_id, so unchanged docs are skipped without any GET-detail call.

First-time init

When the user runs sync/pull/status and no state file exists, the script auto-detects the local layout from .md files under --root (default cwd):

• flat — all .md files at top level. Slug = file stem.
• nested — .md files under subdirectories. Slug = file stem (subdirs are organizational only; they do not create TOC groups).
• frontmatter — every file has --- slug: xxx --- frontmatter at the top. Slug = frontmatter value.
• empty — no .md files yet; suggest pull --all to seed locally.

If layouts are mixed or partially-frontmatter, the script errors and asks the user to pass --layout flat|nested|frontmatter or normalize the directory.

Frontmatter fields (frontmatter layout)

In frontmatter layout, each .md file begins with a YAML block. Supported fields:

• slug (required) — the Yuque document slug (URL identifier). Usually English kebab-case, e.g. create-draft-task.
• title (optional) — the document title pushed to Yuque. Useful when the file name or H1 should not be used as the title.

Example:

---
slug: create-draft-task
title: 创建起草任务
---

# 创建起草任务
正文内容。

The YAML frontmatter is stripped before pushing to Yuque — it never appears in the rendered document body.

Title extraction priority

When sync pushes a document, the title sent to Yuque is resolved in this order:

1. Frontmatter title field (if present)
2. First H1 heading (# Heading) in the body
3. File name stem (without .md extension)
4. Slug (last resort)

When the title falls back to the file name or slug (steps 3–4), sync --check / status lists these docs under a "Title fallback" section so you know to add an H1 or frontmatter title.

Use sync --force-title to always use the file name stem as the title, ignoring H1 and frontmatter title. This is useful when upstream-generated titles are unreliable.

Automatic body transformations on push

During sync push, the following transformations are applied to the body before sending it to Yuque:

1. Frontmatter stripping — YAML frontmatter is removed (frontmatter layout only).
2. Inter-document link rewriting — [text](file.md) links are converted to [text](slug) using the local slug map, so they resolve correctly on Yuque. External URLs (http://...), anchors (#section), and non-.md links are left untouched. Unresolved links are kept as-is and reported as warnings.
3. Bold format fix — **标签：**值 (bold-close marker directly followed by a non-space character) is rewritten to **标签：** 值, because Yuque's renderer requires a trailing space to recognize the bold-close marker. Already-correct **标签：** 值 is not affected.

On pull, the reverse transformation is applied: slug links are converted back to local file-name links when a matching local file is known.

Sync flow per run

1. Lists all remote docs once (with optional_properties=latest_version_id) — no per-doc GET.
2. For each local file, compares normalized SHA-256 of the body and the remote latest_version_id to the values stored in .yuque-sync.json.
3. Pushes:
   • create for local files with no remote counterpart (auto-adds to TOC).
   • update for local files whose body changed and remote did not.
   • Skips unchanged, conflicts (both sides changed, or remote moved), and non-markdown remotes (with a warning).
4. After push, if any tracked doc has gone missing locally, stops with exit code 2 and prints a confirmation block (see below). Never auto-deletes remote docs.

Confirmation handling (exit code 2)

sync exits 2 when local files that were previously synced are now missing. This is not an error — it means the script needs an explicit choice from the user. The stdout block describes three options, each with the exact flag to re-run with:

• --on-missing delete — delete on Yuque too
• --on-missing pull — restore locally from Yuque
• --on-missing forget — keep on Yuque, drop from local state

When you see exit code 2, show the printed block to the user (translated to their query language; see Language section), then re-invoke sync --on-missing once they pick.

Conflicts

If a doc was changed both locally and remotely (or remote moved while local stayed), sync lists it under "Conflicts" and skips it. Resolve manually:

• pull --slug --overwrite — accept remote version locally.
• update -f path/to/file.md — overwrite remote with local version.

After resolution, the next sync run will reconcile the state file.

Language

The CLI emits English in fixed-format blocks for stability. When relaying CLI output to the user — especially the sync confirmation block, status sections, and error messages — translate the human-readable text into the user's query language. Keep the following literal and untranslated:

• CLI flag names and values (e.g. --on-missing delete, --layout flat).
• Doc slugs, IDs, file paths.
• Section keys in JSON output.

Failure handling

• Missing .env or token: Stop. Run the first-time setup flow above. Do NOT ask the user to manually edit .env.
• Invalid YUQUE_REPO format: Stop. Re-run setup with the correct URL.
• .gitignore missing .env entry: Warn the user immediately. Do not proceed with any API calls until .env is protected.
• 401 Unauthorized: Stop. Token is invalid or expired — ask the user to regenerate it.
• 404 Not Found: Stop. The repo path or doc ID/slug is wrong — ask the user to verify.
• 429 Rate Limited: Wait briefly and retry once. If it fails again, stop and inform the user.
• No fields provided on update: Stop. At least one of --title, --body, --body-file, --slug, --format, --public must be given.
• Delete without --confirm: The script refuses. This is intentional — always pass --confirm.
• sync exit code 2: Not an error. Confirmation required for missing-local files. Show the block to the user, get their choice, re-run with --on-missing .
• Layout ambiguous on first sync: Show the script's suggestion to the user; do NOT pick --layout for them — ask which structure they intend.

Do NOT guess doc IDs or slugs. If unknown, run list, toc, or status first to find the target.

Resource navigation

• scripts/yuque_cli.py — the CLI tool. Run it; do not read it unless debugging.
• scripts/yuque_cli.py --help and scripts/yuque_cli.py --help for full flag details.
• The project root .env file holds credentials (never commit it).

High-value pitfalls

1. Creating a document does NOT auto-add it to the TOC. The script handles this by default via append_doc_to_toc. Use --no-toc only if you explicitly want an unlisted document.
2. The id parameter in get/update/delete can be either an integer doc ID or a string slug. Both work, but slugs are more readable.
3. Body content with shell-special characters: Use --body-file instead of --body for any non-trivial content to avoid shell quoting issues.
4. The Yuque Lake format is proprietary. Always use --format markdown (the default) unless the user specifically needs Lake format.
5. TOC insertion always appends to the repo root as a child node. There is no way to specify a parent node or insert at a specific position via this skill. If the user needs custom TOC structure, they must reorder it manually in the Yuque UI.
6. sync ignores remote drafts (body_draft). Change detection is based on latest_version_id (the last published version). If a teammate is editing in the Yuque web editor without publishing, that draft will be overwritten by sync update. If you suspect this, use pull --slug --overwrite to inspect remote first.
7. Never auto-delete remote docs. sync will only delete remotely when the user explicitly passes --on-missing delete after seeing the confirmation block. Do not pass delete on the user's behalf without an unambiguous yes.