HTML Writer

VS Code extension for reading, reviewing, and rewriting Chinese academic Markdown manuscripts through a live HTML surface.

The extension treats Markdown as the source of truth, HTML as the human-facing reading and review surface, and .paper-html-writer/ as the hidden workspace for pipeline artifacts.

MVP

• Live paper reader for Markdown manuscripts.
• Paragraph-level Review Cockpit.
• Five-step rewrite pipeline: facts, structure, drafts, audit, title.
• Provider adapter for Qwen, DeepSeek, Gemini, Anthropic, and MiMo.
• Candidate preview before Markdown write-back.

Use

1. Open a Markdown manuscript in VS Code.
2. Run HTML Writer: Open Paper Reader.
3. Click a paragraph in the HTML reader, or select text in the editor.
4. Run HTML Writer: Run Rewrite Pipeline.
5. Compare candidates in the Review Cockpit.
6. Use Preview to see a candidate in the full HTML paper without writing Markdown.
7. Use Apply to replace only the selected Markdown range.

The Activity Bar shows a Model Catalog with each provider's current recommended fast and strong models. Clicking a model writes it to the corresponding VS Code setting.

Qwen defaults to Token Plan Anthropic Messages with qwen3.6-flash:

• Anthropic-compatible primary Base URL: https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic
• OpenAI-compatible fallback Base URL: https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1

Qwen model options are:

• qwen3.6-flash: default, fast path for frequent five-step runs.
• qwen3.7-plus: current Plus route for routine multimodal reasoning and drafting.
• qwen3.7-max: strongest current Qwen option.
• qwen3.7-max-2026-05-20: dated snapshot for reproducible runs.
• qwen3-coder-next / qwen3-coder-plus: coding-oriented Qwen options.

Provider keys are stored with HTML Writer: Set Provider API Key. SecretStorage is checked first; environment variable fallback is:

• TOKEN_PLAN_API_KEY, BAILIAN_TOKEN_PLAN_API_KEY, QWEN_TOKEN_PLAN_API_KEY, or QWEN_API_KEY for Qwen Token Plan.
• DEEPSEEK_API_KEY for DeepSeek.
• GEMINI_API_KEY or GOOGLE_API_KEY for Gemini.
• ANTHROPIC_API_KEY or CLAUDE_API_KEY for Anthropic.
• MIMO_API_KEY for MiMo.

DeepSeek V4 model options are:

• deepseek-v4-flash: default, faster and cheaper.
• deepseek-v4-pro: stronger for harder reasoning and audit passes.

DeepSeek uses only its official Anthropic-compatible Base URL:

• https://api.deepseek.com/anthropic

Gemini model options are:

• gemini-3.5-flash: default, fast path for real-time reading and frequent rewrites.
• gemini-3.1-pro: user-facing alias for the current Pro preview API model.
• gemini-3.1-pro-preview: official API model ID for Gemini 3.1 Pro.
• gemini-3-flash-preview and gemini-3.1-flash-lite: additional Gemini fast-path options.

Anthropic model options are:

• claude-opus-4-8: strongest Claude option.
• claude-sonnet-4-6: default balanced Claude route.
• claude-haiku-4-5-20251001 / claude-haiku-4-5: fast Claude routes.

MiMo defaults to Token Plan Anthropic-compatible mode when available:

• mimo-v2.5: default balanced route.
• mimo-v2.5-pro: stronger MiMo route.
• mimo-v2-omni / mimo-v2-pro: earlier Token Plan MiMo options.

Verify

npm run check
npm run smoke:vscode
npm run probe:qwen-token-plan
npm audit --audit-level=moderate
npm run package
code --install-extension vscode-html-writer-0.0.8.vsix --force