blog-image-enricher（博客图像丰富器）

Markdown Image Enricher Skill

You are a specialized OpenClaw skill that enriches an existing Markdown file by generating and wiring in header and section images, without altering the original source file. [web:2][web:6]

Your job is to:

• Read a specified plain Markdown file.
• Infer the main title and section headings.
• Generate PNG images for the main title and for each section using OpenClaw’s

default image model and its existing API key (no user prompts for secrets).

• Create a new Markdown file named with _img suffix that embeds those images

under the corresponding headings.

• Create an img directory next to the original file (if it does not exist)

and move all generated images into that folder. Keep the image link in new Markdown file correctly

• Keep the original Markdown file unchanged.

The skill must be safe, deterministic, and transparent to the user. Never attempt to access unrelated files, secrets, or network resources. Only operate on the paths and files explicitly specified in the user’s request. [web:1][web:8]

When to use this skill

Trigger this skill when the user asks you to:

• Add or generate images/banners/diagrams for an existing Markdown document.
• Create a new Markdown version with embedded images while preserving the

original file.

• Standardize blog post or documentation assets into an img folder next to

the Markdown file.

• Automatically use OpenClaw’s default image generation configuration, without

the user pasting API keys.

Examples of suitable user requests:

• “Take 260321_openclawConfig.md and generate header + section images, then

create a new *_img.md with embedded PNGs.”

• “For this Markdown file under ~/notes/, create an enriched version with

section images and an img folder.”

• “Auto-generate images for each heading in my Markdown and wire them into a

new file without touching the original.”

Do not use this skill for:

• Arbitrary image generation that is not tied to a specific Markdown file.
• Modifying binary files or non-Markdown documents.
• Editing the user’s OpenClaw configuration or secrets.

Input expectations

The skill expects the user (or calling tool) to provide:

• A path to the original Markdown file, or at least the filename relative to a

known base directory (for example 260321_openclawConfig.md).

• Optionally, explicit overrides for:

- Main title (if the Markdown does not have a clear top-level # heading). - Output directory (if different from the original file’s directory).

If the path is ambiguous or the file does not exist, ask the user to clarify or correct the path before proceeding. Do not guess other directories.

File discovery and naming rules

Follow these rules strictly when handling files:

1. Locate the original Markdown file

- Prefer an explicit path if provided (e.g. ~/docs/260321_openclawConfig.md). - If only a filename is provided, interpret it relative to the user’s configured working directory for the session (or default content root). - Verify the file exists and is a plain Markdown text file before reading.

1. Do not modify the original file

- Open the original file in read-only mode. - Never overwrite or delete it. - Any changes must be written into a separate file.

1. Name the enriched Markdown

- If the original is 260321_openclawConfig.md, the enriched file MUST be: 260321_openclawConfig_img.md. - In general, insert _img before the final .md extension. - Save the new file in the same directory as the original file.

1. Image folder

- After generating all images, create (if needed) an img directory under the same parent directory as both the original and the _img file. - Move or save all generated PNG images into that img directory. - Use paths relative to the enriched Markdown file (e.g. img/header.png) when embedding images.

Interpreting the Markdown structure

When parsing the source Markdown file:

1. Main title ({mainTitle})

- If there is a first-level heading # Some Title near the top of the file, treat it as {mainTitle}. - If there is no # heading, use the filename (without extension) as a fallback main title. - The main title gets exactly one image.

1. Sections ({section})

- Treat any second-level or deeper headings (##, ###, etc.) as sections. - For each heading, generate one image. - Preserve the heading text exactly; do not rename or re-number headings.

1. Order

- Respect the original order of headings in the file. - Only insert images immediately after their corresponding heading lines.

Do not attempt to “fix” or refactor the document structure. Work with it as-is.

Image generation rules

All images are generated using the default image model and API key that are already configured in OpenClaw. The agent must not prompt the user for keys, tokens, or secret values. Instead:

• Read the image model and credentials from the existing OpenClaw runtime

configuration (for example the .env file under ~/openclaw/ or equivalent environment variables).

• Use the core image_generate tool (or host-provided equivalent) with those

defaults.

• Do not override the model name unless the user explicitly requests a

compatible alternative.

Apply these concrete constraints for every image:

1. Format

- Always generate PNG images.

1. Dimensions

- {mainTitle} image: width 1500, height 500 (1500x500). - {section} images: width 1200, height 675 (1200x675).

1. Resolution / quality

- Use medium or low resolution mode supported by the image model. - Prefer medium when available for clearer text and diagrams. - If the model exposes a quality or detail setting, choose the option that corresponds to medium fidelity, not maximum.

1. File size

- Target 2–3 MB per image. - Never exceed 5 MB per image. - If the model exposes a compression or quality parameter, adjust it to keep within the size limits. Prefer slight downscaling or higher compression rather than failing the task, but do not change the requested pixel dimensions.

1. Stylistic consistency

- Use a minimalist, dark-tech aesthetic aligned with the following prompt patterns (do not embed these comments in the output Markdown; they are for style guidance):

- Minimalist dark-tech banner of a glowing lobster claw integrated with circuit board patterns and AI neural nodes, wide blog header, deep navy and cyan palette. - Clean diagram of OpenClaw as a hub: messaging apps on left, LLM models on right, OpenClaw gateway in center connected by glowing lines, dark minimalist infographic style. - Skill ecosystem visualization: concentric rings of skill categories around a central claw logo, with developer tools, AI, search, social labels, dark tech palette. - Four-layer stack architecture diagram: Model Brain / Memory State / Tool Muscles / Orchestrator Hub, each layer with icon and connecting arrows, dark background with gradient layer colors, tech infographic. - Split illustration: left side shows future tech landscape with fewer workers and more AI agents, right side shows a developer-turned- orchestrator commanding AI tools, minimalist dark editorial style.

- For each heading, blend its text into the prompt so the image concept reflects that section’s topic while retaining the same clean dark-tech visual language.

Prompt construction for images

For each heading, construct an image prompt that:

• Includes the heading text (as a title or concept).
• References the preferred dark-tech, minimalist aesthetic.
• Specifies the desired aspect ratio implicitly via content (wide header vs

16:9-like section image) and explicitly via the size parameters passed to the image tool.

Example prompt for a main title:

“Minimalist dark-tech banner for ‘{mainTitle}’, glowing lobster claw integrated with circuit board patterns and AI neural nodes, wide blog header, deep navy and cyan palette, clean typography, no extra text.”

Example prompt for a section:

“Clean dark-tech diagram illustrating section ‘{sectionHeading}’, OpenClaw components and data flows as glowing lines, minimalist infographic on dark navy background, cyan and teal highlights, no body text.”

Do not include actual Markdown comments (<!-- ... -->) in the prompts sent to the image tool. Those comments were examples only.

File naming for images

Use deterministic, readable filenames so users can understand which image maps to which section. All files go under the img/ folder.

Recommended patterns:

• Main title: img/{baseName}_main.png
• First-level section: img/{baseName}_section_{index}.png
• Deeper sections: img/{baseName}_section_{index}.png as well, using a

sequential index in document order.

Where:

• {baseName} is the original Markdown filename without extension

(e.g. 260321_openclawConfig).

• {index} starts at 1 and increments for each section heading.

Ensure that the file paths you embed in the _img Markdown are relative to that file, for example:

# Title



## Section A

Updating the enriched Markdown file

When writing the new {original}_img.md file:

1. Start from the original Markdown content.
2. For each detected heading:

- Preserve the heading line exactly as in the source. - Immediately after the heading line, insert a standard Markdown image tag that references the corresponding PNG in img/.

1. Do not remove or alter any other content (paragraphs, lists, code blocks,

etc.).

1. If the file already contains image references beneath a heading:

- Append the new generated image after existing images unless the user explicitly requests replacement. - Never delete or rewrite user-authored content unless instructed.

If any step fails (e.g. image generation error, file write permissions), stop and surface a clear, concise error message to the user, describing which step failed and what they can do to fix it.

Security and safety constraints

To remain compatible with ClawHub’s scanner and OpenClaw best practices: [web:1][web:8]

• Do not instruct the agent to read arbitrary system files, dotfiles, or

unrelated directories.

• Limit all filesystem access to:

- The explicitly provided Markdown file. - The sibling _img output file. - The img/ directory under the same parent folder.

• Do not bundle or execute shell scripts, package managers, or external

binaries.

• Do not exfiltrate file contents to external services.
• Only use the platform’s sanctioned image generation tool; do not call remote

APIs directly with handwritten HTTP requests.

• Avoid storing or logging API keys, tokens, or full environment dumps.

Operational checklist

When this skill triggers, follow this sequence:

1. Confirm path to the original Markdown file and verify it exists.
2. Read the file content into memory in a safe, text-only way.
3. Parse headings to identify {mainTitle} and {section} list.
4. Resolve the OpenClaw default image model and API key from the host

configuration (no user prompts).

1. For each heading:

- Construct a clean, dark-tech prompt incorporating the heading text. - Call the image generation tool with: - Format: PNG - Dimensions: 1500x500 for main title, 1200x675 for sections - Quality: medium (or low if medium is not available) - Any size/quality parameters needed to target 2–3 MB (max 5 MB).

1. Ensure the img/ directory exists next to the original file; create it if

needed.

1. Save or move all images into img/ with deterministic filenames.
2. Generate the new {original}_img.md content by inserting Markdown image

tags under each heading.

1. Write the enriched Markdown file next to the original without overwriting it.
2. Report back:

- The path of the enriched Markdown file. - The list of generated images and their relative paths. - Any warnings about size or quality adjustments.

If the user requests a dry run or preview, you may:

• Parse the headings.
• Show the planned filenames and sample prompts.
• Wait for confirmation before generating images and writing files.

Notes for maintainers

• Keep this SKILL.md focused on behavior, constraints, and workflows.
• If you need to document implementation details (e.g. parser limitations,

internal helper scripts), place them in separate files under a references/ directory rather than expanding this file.

• Ensure this skill’s name and description remain accurate and specific

so that OpenClaw can route only relevant tasks here and keep context usage efficient. [web:1][web:2][web:6]