wiki-entity-upgrade

name	wiki-entity-upgrade
description	Upgrade bio-only blogger entity pages to comprehensive thought analysis format
category	wiki

Wiki Entity Page Upgrade

Upgrade bio-only entity pages to comprehensive thought analysis format following mitchellh-com.md as reference.

Scope

This skill covers upgrading entity pages from bio-only to comprehensive analysis. For newsletter triage ingestion (creating new pages from ChinAI/newsletter "take" decisions), see the cn-media-analysis skill and the standard ingest pipeline instead.

Workflow

1. Read existing bio-only entity page in ~/wiki/entities/
2. Scrape author's blog (homepage + 3-5 recent articles)
3. Extract core ideas, philosophy, key quotes, recent themes
4. Write upgraded page following the format below
5. Update wiki/index.md and wiki/log.md
6. Commit: cd ~/ai-topics-cn && git add wiki/ && git commit -m "wiki: upgrade <slug> to thought analysis" && git push

Page Format (Blog Authors / Thought Analysis)

---
title: "Author Name"
type: entity
url: "blog.example.com"
category: blog-author
tags: [topic1, topic2]
---

| Field | Value |
|-------|-------|
| Name | Author Name |
| URL | blog.example.com |
| Known For | Key achievement |
| Themes | Primary writing topics |

## Overview
...

Page Format (Company/Product Entities)

Company and product entity pages (e.g., modelbest.md, doubao.md) use a different structure:

---
title: "Company Name (Chinese) — Brief descriptor"
created: YYYY-MM-DD
updated: YYYY-MM-DD
tags: [company, ai, llm, china, ...]
aliases: ["Chinese", "English", "Abbrev"]
source_lang: zh-CN
---

# Company Name (Chinese) — Brief descriptor

> **Key stat**: value
> **重要度**: 高/中/低 — one-line significance

## 概要
2-3 paragraph overview in Japanese

## Market Data / 市場データ
| 指標 | 値 | 出典 |
...

## 開発歴史 / Timeline
| 時期 | マイルストーン |
...

## 技術スタック
### Subsection per technology area

## 商用展開
Business applications and partnerships

## 競合比較
| 企業 | 路線 | 特徴 |
...

## 関連
- [[wikilink]] — connection explanation

## Sources
- URLs

Key differences from thought analysis format:

• Uses Japanese section headers (概要, 技術スタック, 商用展開, 競合比較)
• Market data and timeline tables instead of quotes
• Competitor comparison tables
• No "Core Ideas" / "Key Quotes" / "Recent Themes" sections
• aliases array includes both Chinese and English names
• tags include company, ai, and specific domain tags

Progress Tracking

After each batch, run this audit to separate completed vs remaining:

import os
target = os.path.expanduser("~/wiki/entities/")
skip = {'agibot-10000-units', 'amazon-rivr', 'anthropic', 'claude-mythos', 'cursor-3',
        'gemma-4', 'glm-5-zai', 'glm-5v-turbo', 'meta', 'mistral-voxtral-tts',
        'muse-spark', 'openai-spud', 'qwen3-6-plus', 'zoox-expansion'}
bio_only, has_analysis = [], []
for f in sorted(os.listdir(target)):
    if not f.endswith('.md') or f.replace('.md','') in skip: continue
    with open(os.path.join(target, f)) as fh: content = fh.read()
    if "## Core Ideas" in content or "## Key Themes" in content:
        has_analysis.append(f)
    else:
        bio_only.append(f)
print(f"✅ Done: {len(has_analysis)}, 📋 Remaining: {len(bio_only)}")
for b in bio_only: print(f"  - {b}")

NOTE: This audit script only detects blog-author thought analysis upgrades (which use ## Core Ideas / ## Key Themes). Company/product entity pages (e.g., modelbest.md, doubao.md) use different section headers (## 概要, ## 技術スタック, ## 競合比較) and will appear as "bio_only" even when fully written. For company pages, check manually: if the file is >5KB and has 5+ sections beyond ## 概要, it's complete.

Batch Processing Strategy

• Budget awareness: delegate_task has ~50 iteration budget per subagent. A batch of 5 entities can exhaust it, leaving some files unwritten. Always verify post-batch.
• Group by similar domains (AI researchers, security experts, infrastructure devs, etc.)
• Process 3-5 entities per subagent task using delegate_task
• Use 2-3 parallel subagent tasks per batch (don't exceed 4 — budget compounds)
• After each batch: audit → copy from wrong dir → commit → push → next batch
• For the final ~10 entities: consider running sequentially or in 2 parallel tasks to ensure completion
• Non-entity pages (companies, models, products) use different formats — skip them
• Target: ~69 blogger entities total (OPML has 84 feeds, but 14 are concepts/products)

Critical Pitfalls

• Canonical write target: delegate_task subagents must write directly to ~/wiki/entities/. If a delegated context exposes an isolated HOME or shows ~/.hermes/home/..., treat that as a runtime artifact and continue to target the canonical wiki path only.
• Subagent filename aliasing: Even when given exact filenames, subagents create files with DIFFERENT names (e.g., benjamin-clavi.md instead of bclavie.md, ethan-mollick.md instead of emollick.md, hynek-schlawack.md instead of hynek.md). After each batch:
  1. Check all new files: ls -la ~/wiki/entities/*.md (sort by time)
  2. Look for skeleton duplicates: files with status: skeleton in frontmatter
  3. Delete skeleton duplicates after confirming enriched version exists
  4. Common pattern: full-name-slug.md vs handle.md — both exist for same person
• Subagent budget exhaustion: Each delegate_task subagent has ~50 iteration budget. When processing 5+ entities, the subagent may hit the limit and complete without writing all files. ALWAYS verify:
  1. Check subagent summary for which files were actually written vs just researched
  2. If exit_reason: max_iterations, some files may be missing
  3. Run the progress tracking audit immediately after each batch
• Non-entity page confusion: OPML has 84 feeds but only ~69 are blogger entities. The rest are companies, products, or concepts. Use the skip set in progress tracking to avoid wasting time on these.
• Hillel Wayne newsletter duplication: Hillel Wayne has both hillel-wayne.md and buttondown-com-hillelwayne.md. The newsletter version should cross-reference the main page, not duplicate all content.

X Account Enrichment Workflow

For X/Twitter accounts (from ~/x-accounts.yaml), the enrichment process is similar but with format differences:

X Account Page Format

---
title: "Full Name"
handle: "@twitter_handle"
created: 2026-04-10
updated: 2026-04-10
tags: [person, topic1, topic2]
aliases: ["handle", "alt-name"]
---

# Full Name (@handle)

| | |
|---|---|
| **X** | [@handle](https://x.com/handle) |
| **Blog** | [URL](URL) |
| **GitHub** | [username](https://github.com/username) |
| **Role** | Job title |
| **Known for** | Key contributions |
| **Bio** | 2-3 sentence background |

## Overview
## Core Ideas
## Key Work
## Blog / Recent Posts
## Related People
## X Activity Themes

X Account Enrichment Steps

1. Check ~/x-accounts.yaml for account list
2. Run ~/scripts/build_x_wiki.py to create skeleton pages (if not already done)
3. Prioritize by importance: high-impact AI figures first
4. Batch process 5 accounts per delegate_task subagent
5. Use 3 parallel subagents per batch (don't exceed 4 — budget compounds)
6. After each batch: audit → cleanup duplicates → commit → push → next batch
7. Target quality: 8-15KB per page, matching antirez-com.md or simon-willison.md depth

Known Duplicate Patterns (X Accounts)

• benjamin-clavi.md → keep bclavie.md
• chip-huyen.md → keep chipro.md
• ethan-mollick.md → keep emollick.md
• eugene-yan.md → keep eugeneyan.md
• hynek-schlawack.md → keep hynek.md (or vice versa, check which is enriched)
• lilian-weng.md → keep lilianweng.md
• samuel-colvin.md → keep samuelcolvin.md
• peter-steinberger.md → keep if enriched (no alternate)
• karri-saarinen.md → keep if enriched (no alternate)
• daniel-han.md → keep if enriched (no alternate)
• cl-mentine-fourrier.md → keep clefourrier.md
• late-interaction.md → concept page, not person (delete if in entities/)