Overview

At a glance

• Total pages: 2208
• Checked against sources (AI): 1171 / 1188 (99%) (claude-opus-5, claude-opus-5-5, claude-opus-5-5[1m], claude-opus-5[1m], claude-sonnet-5, claude-sonnet-5-5)
• Findings raised and fixed before publishing: 382
  • Statements the sources do not make: 201
  • Statements that contradict the sources: 74
  • Statements resting on a document not among the sources: 66
  • Other: 41
• Translation pages: 1020 (no check against the original is recorded)
• Pages withheld because they did not pass the check: 1
• Read and checked by a person: 36 / 2208 (2%)
• Types in use: 16
• Pages per language: en 1188, ja 1020
• Translation coverage: ja 1020 / 1188 (86%)

Pages are published as soon as an LLM generates them. The check against sources is run by machine; “read and checked by a person” is how many pages someone has since read all the way through without anything obviously wrong standing out. Only some pages are read by a person, by design — this number is not meant to reach the total, and it is not a complete quality guarantee.

Each page generated from sources is checked against those sources when it is generated. That check covers agreement with those sources and nothing else — not completeness, not the effect on real people and organizations, not conflicts with what you know.

Translation pages are outside this check and are left out of the ratio above.

A page that could not be brought in line with its sources while it was being generated is not published. This counts the check at work, not a defect in the wiki.

Knowledge hubs

The pages other pages link to most.

“Sources recorded” is the number of sources recorded on the page, not the number of independent confirmations. A page about one particular document has exactly one by construction — fewer is not worse.

• Claude Code — Backlinks: 218 / Sources recorded: 9
• Context engineering — Backlinks: 113 / Sources recorded: 7
• Model Context Protocol — Backlinks: 110 / Sources recorded: 6
• Spec-driven development — Backlinks: 105 / Sources recorded: 12
• Vibe Coding — Backlinks: 87 / Sources recorded: 13
• Human-in-the-loop — Backlinks: 82 / Sources recorded: 7
• AGENTS.md — Backlinks: 79 / Sources recorded: 13
• Harness Engineering — Backlinks: 74 / Sources recorded: 10
• OpenAI Codex — Backlinks: 64 / Sources recorded: 4
• Agent Skills — Backlinks: 62 / Sources recorded: 9
• Guardrails — Backlinks: 59 / Sources recorded: 6
• Prompt injection — Backlinks: 54 / Sources recorded: 7
• Anthropic — Backlinks: 54 / Sources recorded: 2
• Compaction — Backlinks: 53 / Sources recorded: 9
• Agent Harness — Backlinks: 51 / Sources recorded: 7
• Cursor — Backlinks: 48 / Sources recorded: 5
• GitHub Copilot — Backlinks: 48 / Sources recorded: 4
• Review Bottleneck — Backlinks: 42 / Sources recorded: 3
• Agentic Code Review — Backlinks: 41 / Sources recorded: 5
• SWE-Bench — Backlinks: 39 / Sources recorded: 4
• and 1020 more

By type

Gaps

Referenced but not written

Pages other pages link to that do not exist yet.

• DefinedTerm/tool-use — Referenced by: AI Coding Agent, Plan-Act-Observe Loop, Sandboxing
• DefinedTerm/ai-assisted-code-review — Referenced by: The 70% Problem
• Organization/alibaba — Referenced by: Open Code Review
• Organization/plandex-ai — Referenced by: Plandex
• Organization/tikal — Referenced by: Agentic SDLC Spec Kit

Pages nothing links to

• A Guide to Fine-Tuning FunctionGemma
• 진짜 AI 에이전트형 코딩 툴 ＇Claude-Code＇
• Год с Claude Code: главное — не он сам, а то, что в .claude/
• Agent 编程方法论：不教工具教方法，让 AI 按你的规矩干活
• Agent pull requests are everywhere. Here’s how to review them.
• Claude Code Agent Skillsを活用したTECH BLOGレビュー ── AIで推進するレビュー自動化
• Agent Skills
• Agentic Code Review
• Agentic Coding mit Spec-Driven Development: Warum Ihr KI-Agent Struktur braucht, bevor er Code schreibt
• Agents CLI in Agent Platform: create to production in one CLI
• AI Assisted Code Reviews - Augmenting Pull Requests in .NET Projects
• AI-assisted Coding과 개발 생산성 향상 (부제: 실리콘밸리 개발자가 바라보는 AI 생산성 툴 및 개발 방식의 변화)
• AI 编程 2025 总结：国产模型“能力追平”，国产编程工具还在“情感陪伴”
• AI Coding Agents: Adoption Trends
• AI 友好架构：DevOps 平台 & 平台工程赋能 AI 自动编程
• The Anatomy of Harness Engineering: How to Evaluate, Iterate, and Guard AI Coding Agents
• Pythonでゼロから作るコーディングエージェント
• 人材紹介開発におけるコーディングエージェントの導入状況や勉強会の様子
• 设计流动摩擦：AI 原生团队的核心能力
• AIエージェント時代の開発 — エージェンティックコーディング実践
• and 193 more

Sources

By type

By status

By host (URL sources)

By language

The language a source’s extracted text is mainly written in, as judged by the LLM at generation time. Pages are always written in this wiki’s primary language, so a row for another language is expected: the pages written from those sources went through a summarizing translation. “Not recorded” counts sources taken in before this was recorded that have not been re-read since. “Not processed yet” counts registered sources that have never finished a run (the queue).

Source type x generated page type

Tags