Agent-Friendly Blog Design

Context

The blog is a Jekyll static site published at https://blog.sunxie.me.

The current site already has:

The user wants to make the site more friendly to AI agents in two distinct ways:

  1. Agents should be able to discover, understand, and cite the site’s content more easily.
  2. Agents should be able to submit article drafts to the blog owner using a very lightweight protocol.

The user explicitly prefers a modern “website with skills” shape over a purely SEO-oriented approach.

The chosen direction is to expose one public website skill that explains:

The submission path is intentionally minimal:

Goals

Non-Goals

Chosen Approach

Use a hybrid but minimal architecture:

This approach is preferred because it serves both classes of agents:

Public Surface

The site will expose three public entry points.

1. Discovery Entry Point

Add /llms.txt as the first machine-readable overview of the site.

Its role is to tell an agent:

This file is for initial orientation, not for the full operating contract.

2. Skill Discovery Entry Point

Add /.well-known/skills/index.json.

Its role is to advertise one website skill, sunxie-blog, so skill-aware agents can discover it directly from the domain.

The index should point to a public SKILL.md file.

3. Website Skill

Add a public skill document at /skills/sunxie-blog/SKILL.md.

This file is the core contract for agent interaction with the site.

It will contain three sections:

Skill Responsibilities

Discover

The Discover section should instruct agents to:

It should also describe the blog at a high level, for example:

Consume

The Consume section should instruct agents to:

This section is not an API specification. It is a reading and citation protocol.

Submit

The Submit section should define the article draft contract.

For the first version, agents must:

Agent Submission Rules

These rules apply only to agent-submitted articles.

They do not apply to all human-written posts on the site.

Allowed Categories

The categories field must contain exactly one of the following:

Required Tag

The tags list must include:

Additional tags are allowed and encouraged when they clarify the topic.

Markdown Template

The skill should provide a concrete template similar to:

---
title: "文章标题"
excerpt: "一句话摘要"
date: 2026-04-12 10:00:00 +0800
categories:
  - AI-Essays
tags:
  - AI-GENERATE
  - AI Agent
---

This should be followed by the full article body in Markdown.

Email Delivery

The submission instructions should state:

The first version does not require an automated acknowledgment or webhook.

Site Content Expectations

To make the skill useful, the website itself should present cleaner machine-readable signals.

The implementation should therefore include:

The goal is not to duplicate the skill everywhere, but to make the website coherent for both humans and agents.

Information Architecture

The proposed file layout is:

If static-site routing constraints make the final public URLs slightly different, the rendered URLs must still match the paths above.

Implementation Notes For Jekyll

The current repository structure suggests the simplest implementation path:

No backend service is required.

No plugin changes are required for the first version unless Jekyll publishing behavior needs a small adjustment for these static files.

Validation

The design is successful when:

Risks And Tradeoffs

These tradeoffs are acceptable for the first version because the user explicitly prefers a lightweight and modern setup over a large workflow system.