LLM Wiki

Appearance

Built 26/04/17 09:08commit f8ff6f9

Index ​

中文 | English

This index is the catalog layer for the vault: it groups durable pages by role and source family so readers can decide where to drill deeper.

Start Here ​

  • Architecture - End-to-end diagram of the current local-first llm-wiki workflow: Web Clipper capture, Codex maintenance, GitHub sync, VitePress publishing, and Vercel deployment.
  • Wiki Overview - Guided entry point for the vault, including reading paths, core topics, and source families.

Concepts ​

  • LLM Wiki - Persistent knowledge pattern that compiles synthesis into markdown and treats raw-path refactors as maintenance rather than re-ingest.

Entities ​

Topics ​

  • Agent-First Repositories - Repository-local docs, tools, and control systems that make autonomous agent work legible and enforceable.
  • Agent-Readable Design Systems - Versioned markdown design contracts that make frontend taste and UI constraints legible to agents.
  • Agent Harness Design - Planner-generator-evaluator patterns, handoff artifacts, and simplification heuristics for long-running agent systems.
  • Codex Operating Practices - Durable operating model for configuring Codex, packaging repeatable workflows, and choosing the right execution surface.
  • Knowledge Operations - Canonical ingest, screenshot OCR, query, lint, and raw-taxonomy-refactor loop for maintaining the bilingual vault and its newest-first log.
  • Vault Architecture - Three-layer model tying together reorganizable raw sources, bilingual indexes, the markdown wiki, and the schema contract.

Source Families ​

Anthropic / Claude Code / Everything Claude Code ​

Anthropic ​

OpenAI / Codex ​

  • OpenAI Harness Engineering In An Agent-First World - Source on Codex-first engineering, repository legibility, and treating docs and tooling as the agent control plane.
  • Codex Best Practices - Source on prompt structure, AGENTS.md usage, testing, MCP, skills, automations, and session controls.
  • Codex CLI Slash Commands - Source on the interactive control surface for live sessions, including permissions, compaction, diff review, and diagnostics.
  • Codex Non-Interactive Mode - Source on codex exec, machine-readable output, CI auth, and non-interactive guardrails.
  • Codex Agent Skills - Source on skill packaging, discovery scopes, implicit invocation boundaries, and plugin distribution.
  • Codex Subagents - Source on delegation limits, custom agent files, and concurrency/depth controls.
  • Codex Plugin For Claude Code - Source on exposing Codex review, rescue delegation, background-job controls, and stop-hook review gates through Claude Code's native plugin surface.
  • GitHub OpenAI Codex Repository - Direct implementation source showing Codex as a multi-crate runtime with explicit app-server, thread, command-exec, approval, and sandbox layers.

OpenAI / Codex / Configuration ​

OpenAI / Codex / Skills ​

  • Codex LLM Wiki Skill - Skill-family source bundle covering the maintainer spec, bilingual rules, automation model, and safe-refactor helpers for repository-backed markdown wikis.

Karpathy ​

Cocoon AI ​

  • Architecture Diagram Generator - Skill-packaged workflow for turning text architecture descriptions into shareable standalone HTML/SVG system diagrams.

Daniel Breunig ​

EveryInc ​

  • Compound Engineering Plugin - Workflow bundle that turns brainstorming, planning, review, and knowledge compounding into reusable commands across agent clients.

Garry Tan ​

  • gstack - Claude Code environment packaged as a virtual engineering office with reusable specialist-role commands and team bootstrap flows.

Multica AI ​

  • Multica - Open-source managed-agents platform that turns coding agents into assignable teammates with unified runtimes, daemon-backed execution, and team-level skill accumulation.

Piebald AI ​

Ralph ​

  • Ralph Wiggum Loop Technique - Source on simple monolithic agent loops, one-item iterations, repo-persisted memory, and bounded subagent use.
  • Ralph Step-By-Step Guide - Source on the concrete scripts/ralph/ loop, PRD/progress files, commit-after-pass discipline, and fresh-context operation.
  • Ralph GitHub Repository - Source on the concrete snarktank/ralph repository, including the shell loop, packaged skills, plugin metadata, and distribution model.

VoltAgent ​

  • VoltAgent Awesome DESIGN.md - Source on DESIGN.md as an agent-readable design contract and on style-reference catalogs as reusable frontend control surfaces.

Shanraisshan ​

Answers ​

  • How To Make Agents Run Longer - Practical checklist for improving long-running agent reliability through harness design, repo legibility, and control loops.

Operations ​