Curated Content Writer

This skill converts external content (papers, blogs, articles, reports) into compact curated entries for the Personal Blog curated section (src/content/curated/).

The goal is not to write a deep research card. The goal is to create a lightweight, scannable recommendation that shows on the homepage.

Curated vs Archive distinction:

• Archive (src/content/archive/): deep notes, research fragments, knowledge cards. Long-form, reference-oriented.
• Curated (src/content/curated/):精选推荐, concise actionable summary. Short-form, homepage-oriented.
• A curated entry can link to an archive card via relatedArchive when there's deeper context.

Primary target

Write into:

• /mnt/hermes-data/personal/projects/blog/src/content/curated/

Workflow

When triggered, do this sequence:

1. Read the user's input carefully — they'll typically send a URL.
2. If the input includes a URL, fetch the actual content first before writing.
   • For papers: use web_extract (works with arXiv PDFs), also check code repos
   • For blog posts: use web_extract or browser-based retrieval
   • For Claude/ChatGPT share links: use Playwright + system Chromium (/usr/bin/chromium-browser) — these are JS-rendered SPAs
3. Read and understand the content. Extract:
   • Core idea (1-2 sentences)
   • Why it's notable (the insight, not just the topic)
   • How it connects to the user's projects (Hermes Agent, Atypica, etc.)
   • Check code repos (for papers): is there a valid repo? What's the code quality like?
4. Decide whether a matching curated entry already exists. If so, update it.
5. Write the curated entry file.

File naming

Use kebab-case English slugs. No date prefix needed (the date field in frontmatter handles ordering).

Examples:

• hermes-fts5-session-search.md
• prompt-caching-engineering.md
• claude-html-effectiveness.md

Curated schema

---
title:              # required — your title (concise, descriptive)
description:        # required — 1-2 sentence "why this matters" (max 200 chars)
date:               # required — current date
updatedDate:        # optional
source:             # required — URL to the original paper/blog/article
sourceTitle:        # optional — original title of the source
sourceAuthor:       # optional — author / organization name
tags:               # optional — reuse existing tags
type:               # required — enum: paper | blog | article | report
status:             # optional — enum: curated | digested. Use 'digested' if you've read and analyzed it deeply.
relatedArchive:     # optional — list of related archive card slugs (without .md)
heroImage:          # optional — if there's a good OG image
draft: false        # optional
---

type values

• paper — academic papers (arXiv, conference proceedings)
• blog — blog posts (company blogs, personal blogs)
• article — long-form articles, documentation, tutorials
• report — technical reports, white papers

description guidance

Write a compelling 1-2 sentence why-this-matters summary. Not a factual abstract — a judgement call. It should make someone want to click and read more.

Examples:

• ❌ "This paper discusses memory-augmented LLMs" (boring, factual)
• ✅ "Hermes 用 SQLite FTS5 + LLM 做会话搜索替代向量检索——单用户场景下比 vector RAG 更便宜、更稳、零运维" (opinionated, specific)

Body structure

Keep the body concise but substantive. Recommended format:

[Key insight in 1-2 sentences]

**为什么值得看**：[Why is this notable? What makes it different?]

**可落地**：[If applicable, how could this be used in the user's projects?]

The body should be 3-8 sentences total. Use Chinese for Chinese users, English for English users.

Verification

After writing:

1. bun run check (or bun run build) — but only if the build currently passes without errors; skip if pre-existing errors exist
2. Create a git branch → commit → push → PR → merge
3. Check Vercel deployment status

Related skill

Use the archive-card-writer skill when the user provides deep/nuanced content that deserves a full archive card alongside the curated entry.