Convert Document To Markdown

Use this skill when a user wants a supported local file converted into Markdown for later processing.

What this skill does

• - Converts supported local files into Markdown:

• - Image handling modes are file-type dependent:

• - Only runs through Docker. Do not use local Python execution as an operational path.
• Uses a prebuilt Aliyun CR image with fixed version 0.0.1:

• - Returns structured JSON by default so later tool calls can consume markdown, logs, and meta.
• Reads one-time VL configuration from OpenClaw skill config or the repository .env file, then forwards it into the container automatically.
• Only exposes the file command. URL, health, and version commands are intentionally removed to keep startup lean.
• Do not use latest, do not build a fallback image at runtime, and do not treat .doc, .ppt, .xls, audio files, or unlisted image formats as supported inputs.

Required workflow

1. 1. By default the scripts use crpi-4auaoyyj6r36p6lb.cn-hangzhou.personal.cr.aliyuncs.com/huozige_lab.
2. Let the wrapper script resolve the host architecture and choose convert-document-to-markdown-arm64:0.0.1 or convert-document-to-markdown-x64:0.0.1.
3. If needed, override with IMAGE_REGISTRY or IMAGE_NAME.
4. For a local file, run:

1. 5. Parse the JSON result.
2. If success is false, surface error.message and relevant logs.
3. If success is true, use markdown as the canonical output for downstream work.

One-time VL configuration

This skill is designed so the user does not need to re-enter Vision API settings on each run.

Preferred OpenClaw configuration in ~/.openclaw/openclaw.json:

CODEBLOCK0

This works because:

• - skillKey is INLINECODE51
• INLINECODE52 is VL_API_KEY, so apiKey maps to INLINECODE55
• INLINECODE56 can hold VL_BASE_URL and INLINECODE58

Repository-local runtime configuration:

• - copy .env.example to INLINECODE60
• fill VL_BASE_URL, VL_API_KEY, and INLINECODE63
• by default the scripts use INLINECODE64
• optionally override with IMAGE_REGISTRY or INLINECODE66
• use scripts/run_docker_cli.sh, which loads .env, forwards any host VL_* variables into docker run, and pulls the correct fixed-version image if missing

Command patterns

Local file:

CODEBLOCK1

Parameters

• - INLINECODE71

• - INLINECODE72

• - INLINECODE74

• - INLINECODE75

• - INLINECODE76

• - INLINECODE78

• - INLINECODE80

Operational notes

• - For very large local files, stay with the Docker CLI path; do not wrap the file content into base64 or a temporary HTTP service.
• The skill is Docker-only. Do not instruct users to run uv, python, or any other local runtime path for production use.
• The wrapper scripts choose the image by host architecture. Override with IMAGE_ARCH only when you have a concrete reason.
• Prefer IMAGE_REGISTRY plus the fixed version 0.0.1; only use IMAGE_NAME when you need to pass the full image reference explicitly.
• When the user asks for VL or VL-page, first check whether VL_BASE_URL, VL_API_KEY, and VL_MODEL are already configured via OpenClaw skill config or .env.
• If the user only needs extracted Markdown and not the raw JSON wrapper, read the JSON and return the markdown field.
• If the user provides an unsupported extension such as .doc, .ppt, .xls, .wav, .mp3, .m4a, or .mp4, say the current skill does not reliably support it.

Safety notes

• - Treat file paths as untrusted input. Quote shell arguments correctly.
• Do not claim success unless the command returns success: true.