Skip to content

Master Index

Every wiki article listed alphabetically with a one-line summary. Format: - [[path]] — {one-line summary}

  • concepts/agent-harness — The complete software infrastructure wrapping an LLM (orchestration loop, tools, memory, context management, state, error handling, guardrails) that transforms a stateless model into a capable agent; 12 components, framework implementations, and 7 key design decisions
  • concepts/ai-regulation — U.S. and international AI regulation landscape; Colorado SB24-205 DOJ challenge sets precedent for federal preemption of state AI laws; legislative repeal as the path forward
  • concepts/agentic-engineering — Professional AI-assisted software development using coding agents, with emphasis on testing, templates, and engineering judgment; coined by Simon Willison
  • concepts/agentic-workflows — Anthropic's five composable LLM workflow patterns (prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer), sub-agents vs. agent teams distinction, context-based decomposition, and when to use true autonomous agents
  • concepts/ai-red-teaming — Microsoft AIRT's eight lessons from red-teaming 100+ GenAI products: threat model ontology, simple attacks win, PyRIT, XPIA, RAI harms, and why AI security is never complete
  • concepts/context-files — The CLAUDE.md / AGENTS.md / DESIGN.md / SKILL.md family of version-controlled agent context files; persistent project knowledge for AI agents; emerging open convention
  • concepts/frontier-ai-cyber-capabilities — AISI evaluation of 7 frontier models on simulated cyber attacks: Claude Opus 4.6 completed 15.6/32 enterprise attack steps at £65/attempt; 6× improvement in 18 months; NCSC defender advantage framework
  • entities/ethan-mollick — Wharton professor and author of One Useful Thing; known for the Models/Apps/Harnesses framework for understanding agentic AI tools
  • concepts/ai-for-small-business — Frameworks and patterns for integrating LLMs into small business and contracting operations, including data pipelines and automation
  • concepts/ai-inflection-point — November 2025 threshold where coding agents crossed from "mostly works" to "almost always does what you told it" — triggering the dark factory era
  • concepts/claude-code — Anthropic's AI-assisted development tool; enables non-programmers to build apps and professionals to run multiple agents in parallel; updated with official best practices
  • concepts/claude-code-skills — Modular SKILL.md-based capabilities that extend Claude Code; includes bundled skills (/batch, /loop, /simplify) and the agentskills.io open standard
  • concepts/harness-engineering — OpenAI Codex team's model for human engineering in agent-first codebases: designing environments, specifying intent, and building feedback loops; validated by shipping 1M lines of agent-generated code
  • concepts/llm-knowledge-base — Karpathy's pattern for LLM-compiled personal wikis: persistent, compounding artifact vs. RAG; three-layer architecture (raw/wiki/schema); Memex lineage
  • concepts/llm-tier-security — Personal computer security in the AI era: how Mythos-tier exploit capabilities change the threat model and what defenses make sense (network isolation, hardware keys, sandboxing, financial alerts)
  • concepts/mcp-authentication — OAuth 2.1 authentication for MCP servers using Microsoft Entra ID, FastMCP's RemoteAuthProvider, and VS Code pre-registration; includes OBO flow for downstream API access
  • concepts/multi-agent-misalignment — Emergent failure where individually aligned agents collectively produce false institutional records via narrative drift; role-fidelity causes information compression that corrupts organizational state
  • concepts/obsidian-claude-code-os — Obsidian + Claude Code as a personal AI operating system: vault-as-context, Obsidian CLI for relationship graphs, custom slash commands, and inline delegation
  • concepts/openclaw-security — CVE-2026-33579 privilege escalation, OpenClaw's inherent lethal-trifecta risk profile, and guidance to assume compromise for exposed instances
  • concepts/personal-knowledge-management — Atomic notes, personal knowledge graphs, and AI skills: the pipeline from analog reading to active tools via Obsidian
  • concepts/prompt-injection — The primary unsolved security vulnerability in LLM-powered agents; attacker-controlled text overrides developer instructions; lethal trifecta is the most dangerous form
  • entities/andrej-karpathy — AI researcher who coined "vibe coding" and described the LLM knowledge base pattern
  • entities/google-stitch — Google team behind the DESIGN.md open specification for version-controlled agent-readable design rules and conventions
  • entities/harmeet-dhillon — DOJ Assistant Attorney General who announced and led the constitutional challenge to Colorado SB24-205 AI regulation
  • entities/rohit-krishnan — Researcher and writer who demonstrated multi-agent narrative drift in the Helios Field Services experiment using the Vei simulator
  • entities/vannevar-bush — Engineer who described the Memex (1945), a theoretical personal knowledge machine; cited as historical antecedent to LLM knowledge bases
  • entities/pamela-fox — Python/Azure developer and technical educator; author of guide on authenticated MCP server development with Entra ID
  • entities/chuck-kyle — Contractor-turned-digital-marketer and AI early adopter; primary source on practical AI adoption for small business
  • entities/simon-willison — Django co-creator, coined "prompt injection," creator of Datasette; leading practitioner and commentator on agentic engineering
  • entities/vei — Virtual enterprise simulation framework by Rohit Krishnan's Strange Lab; provides persistent company state, role-bounded agents, and replayable seeds for multi-agent research
  • guides/ai-tool-selection — Framework for choosing AI tools in the agentic era: Models/Apps/Harnesses layers, leading models compared, tool recommendations by use case and user level
  • guides/progressive-web-scraping — Four-tier escalating web scraper for Claude Code: WebFetch → cURL → Playwright → Bright Data MCP; Tiers 1–3 free, handles ~95% of sites at zero cost
  • guides/local-agent-stack — Five-layer stack for fully local autonomous agents: llama.cpp, GGUF quantization, llama-server, ChromaDB, LangGraph; zero API costs, full data sovereignty
  • guides/openclaw-docker — Step-by-step guide to running OpenClaw safely in Docker with Telegram integration and web UI access