Dark Factory Kb
Compiled knowledge base — 204 articles
Sources
"About"
A compiled knowledge base documenting the three major autonomous AI agent factory systems:
"Agentic Engineering Patterns"
tags: [agentic-engineering, coding-agents, patterns, llm, tdd, simon-w]
"All Pages — Dark Factory KB"
"Angelica Parente (@draparente) — tufte-viz skill goes viral on X"
> "Realized I created a repo to share this a few months ago, just pushed a couple updates and had Claude generate some pre/post comparisons. Apparently the one day I don't check the internet this goes…
"Appendix A: Tool Reference"
This appendix provides a complete reference for every OpenClaw tool available to practitioners. Whether you're debugging a tool call, designing a new automation, or writing skill definitions — this is…
"Appendix B: Skill Library"
This appendix covers the skill pattern — how OpenClaw capabilities can be extended through packaged tool configurations, prompts, and integrations installed from ClawHub or defined directly in your co…
"Appendix C: Resources"
This appendix pulls together the official resources, tools, and recommended reading that support serious automation work with OpenClaw. It's the practical reference for practitioners who need to go de…
"Appendix D: Automation Cookbook"
This appendix is organized differently from the rest of the book — no theory, minimal explanation, just working recipes you can copy, adapt, and run. Each recipe follows a Problem/Ingredients/Recipe/V…
"Appendix E: Common Patterns Reference"
This appendix serves as a compressed quick reference for the patterns used throughout the handbook. When you're mid-implementation and need a fast reminder of how something works — the SOUL.md structu…
"Appendix F: The Practitioner's Field Guide"
This appendix is organized around questions that come up during real implementation work — the "I'm in the middle of building this and I'm stuck" questions that theory chapters don't answer. It covers…
"Appendix G: Full Chapter Expansions"
This appendix provides deep expansions on topics introduced in the main chapters — the "go further" material for practitioners who want to understand the why behind the patterns. It covers AI model in…
"Appendix H: Extended Troubleshooting Reference"
This appendix is a comprehensive troubleshooting index covering every significant failure mode in OpenClaw automation — organized by system component and searchable by symptom. Rather than theoretical…
"Appendix I: Glossary"
This glossary provides precise definitions for the core terminology used throughout the Kelly handbook and OpenClaw ecosystem. It serves as a quick reference for the canonical meaning of terms as used…
"Appendix J: Version Notes"
This appendix documents what was current when this handbook was written (early 2026) and what practitioners should verify against current documentation as the platform evolves. OpenClaw is actively de…
"Appendix K: Complete Worked Examples"
This appendix contains four complete, end-to-end worked examples, each showing every file, script, and configuration needed to run a specific automation from scratch. Unlike the cookbook which provide…
"Appendix L: Rapid Reference"
This appendix is the 50-most-useful commands and patterns quick reference — the condensed cheat sheet for practitioners who know the concepts but need to recall exact syntax. It covers gateway managem…
"Audit, Test, Automate: How We Decide What AI Can Own"
A task is suitable for AI delegation when all three conditions hold:
"Beads vs Kelly Pipeline Assessment — Can Beads Replace Kelly's State Tracking?"
This assessment evaluates how practical Steve Yegge's Beads framework would be as a replacement for the Kelly pipeline's current state tracking infrastructure. Beads (git-versioned, SQL-queryable work…
"Beads — The Git-Versioned, SQL-Queryable Work Primitive"
A Bead is the atomic unit of work in the Gas Town / Gas City ecosystem. Every discrete unit of work — a code task, a message, a coordination signal, a patrol route, a quality gate — is a Bead. Beads a…
"Chapter 0: Getting Started"
OpenClaw is an AI-powered automation platform that runs as a long-lived Gateway daemon on your machine, exposing AI agents that can read files, execute shell commands, control browsers, send messages,…
"Chapter 10: Browser Agent Deep Dive"
The browser tool is OpenClaw's heaviest but most powerful web automation capability, enabling automation that literally nothing else can do. The Chrome Extension Relay is the key differentiator: inste…
"Chapter 11: Business Automation"
This chapter covers the real production systems that justify OpenClaw as a business tool. The software factory pattern (the Kelly Router architecture from AGENTS.md) takes product ideas through six st…
"Chapter 12: Creative Workflows"
Creative work has a dirty secret: most creative blocks are logistical, not creative. The actual creative work—making it good—is human. But gathering references, organizing research, reformatting conte…
"Chapter 13: Personal Life Automation"
Personal life automation is the most compelling use case because it's your own time and quality of life on the line. Morning briefings are the flagship personal automation—compiling everything you nee…
"Chapter 14: Designing Your Stack"
Designing an automation system that actually works requires discipline: start small with the one repetitive daily task you hate most, not with a grand multi-phase pipeline. The key insight is the 10x …
"Chapter 15: Troubleshooting & Optimization"
The most common failures in rough order: Gateway not running (nothing works; check with openclaw gateway status and start with openclaw gateway start; prevent with LaunchAgent/systemd autostart), wron…
"Chapter 2: OpenClaw Architecture"
Understanding OpenClaw's architecture demystifies everything else. The system is built as a composable pipeline: You → Channel → Gateway → Agent → Tools → World. The Channel (WhatsApp, desktop, CLI) i…
"Chapter 3: File & Code Automation"
Files are the connective tissue of any automation system—every pipeline stores state between runs, passes data between components, logs activity, and accepts configuration through files. OpenClaw prov…
"Chapter 4: Web Automation"
OpenClaw has two web access modes: lightweight (web_search and web_fetch via Brave API) and heavyweight (full browser control via Chromium). Use web_fetch when the content is in the HTML source or you…
"Chapter 5: Communication Automation"
The message tool is OpenClaw's interface to messaging channels—WhatsApp is the primary integration. Basic sends require a target (phone number with country code or saved contact name) and a message st…
"Chapter 6: Time-Based Automation (Cron)"
Cron is Unix's oldest automation primitive, running since 1975, and OpenClaw builds on it with natural language task descriptions that make scheduling accessible. Cron expressions are five fields: min…
"Chapter 7: Multi-Agent Orchestration"
A single agent has hard limits: a finite context window, sequential-only work, no specialization, and no self-check when it goes wrong. Multi-agent systems address all of these—sub-agents work in para…
"Chapter 8: Memory & Context Management"
Context is the most underrated challenge in AI automation. AI models process information through a context window—current models handle 100K+ tokens but context still has hard limits. When context ove…
"Chapter 9: Node Network"
A "node" in OpenClaw is any device paired with your Gateway—phone, tablet, laptop, remote server—that can execute commands or provide data. The node network extends OpenClaw's reach beyond a single ma…
"Claude + Obsidian have to be illegal"
Primary source: the tweet links to a long-form article at https://x.com/i/article/2041858771123617793 (a Twitter article/Card). Full article text was not retrievable via automated tools due to X/Twitt…
"Claude Code Dynamic Workflows: Parallel Subagent Orchestration"
On May 28, 2026, Anthropic announced dynamic workflows as a research preview in Claude Code. The core idea: Claude writes a JavaScript orchestration script on the fly, then a separate runtime executes…
"Cursor for Everything"
DoltHub argues that Cursor's success in the IDE market — combining a familiar UI with an LLM chat window and version control — can be replicated across any application. The key insight: LLMs make mist…
"Cursor for SQL (Introducing Agent Mode in Dolt Workbench)"
DoltHub officially introduces agent mode in the Dolt Workbench — a Cursor-like agentic chat interface that allows users to safely interact with Dolt, MySQL, and Postgres databases. The article demonst…
"Factory Pattern vs Anthropic Dynamic Workflows: A Comparative Analysis"
On May 28, 2026, Anthropic shipped Dynamic Workflows as a research preview in Claude Code, bundled with Opus 4.8. The core mechanism:
"From Smart Notepad to Something More"
title: "From Smart Notepad to Something More"
"Gap Analysis: dark-factory-kb vs Karpathy LLM Wiki Pattern"
index.html links to 0 concept pages — the concept layer is invisible from main navigation
"Gas City — The SDK for Building Custom Dark Factories"
Gas City is the evolutionary successor to Gas Town. Where Gas Town was a single, opinionated product — an orchestrator for Claude Code clones — Gas City is an SDK for building custom dark factories. I…
"GUPP — Gas Town Universal Propulsion Principle"
> If there is work on your hook, you MUST run it. No yielding, no waiting — relentless execution.
"Kelly Factory — Overview"
The Kelly Factory is a dark factory architecture for software development — a multi-agent production line that processes ideas into shipped software with minimal human intervention. It is the operator…
"Kelly vs Gas Town — Full Gap Analysis"
Steve Yegge's Gas Town system and Kelly's factory methodology are two independent inventions of the same underlying pattern: autonomous multi-agent teams executing structured work pipelines with quali…
"Mayor, Crew, and Polecats — The Three-Tier Agent Hierarchy"
Gas Town's most immediately recognizable architectural decision is its deliberate hierarchy of agent roles,，拒绝 the flat "swarm of identical agents" model that most multi-agent frameworks default to. I…
"MEOW — Molecular Expression of Work"
The name is deliberately playful (Yegge's style), but the concept is serious: in MEOW, everything is Work. Not just code tasks, but:
"Open Knowledge Format (OKF) — Google's Standardized Markdown Knowledge Format"
The Open Knowledge Format (OKF) is an open, vendor-neutral specification for representing knowledge as a directory of markdown files with YAML frontmatter. It formalizes the "LLM wiki" pattern — markd…
"OpenClaw Auto Review Skill"
A structured code review system that runs as a closeout check before commit or ship. This is code review, not approval routing — the output is advisory, never authoritative. Codex is the default engin…
"RevenueCat Hires an AI Agent — The First Agentic AI & Growth Advocate"
On March 5, 2026, RevenueCat — a San Francisco SaaS company managing in-app subscriptions for 30,000+ apps, processing $10B+ in annual volume — posted a job listing that made headlines across tech med…
"RevenueCat Hiring Agentic AI & Growth Advocate (2026)"
title: "RevenueCat Hiring Agentic AI & Growth Advocate (2026)"
"SaaS is kinda dead. We just Codex every internal tool we have"
> "SaaS is kinda dead. we just Codex every internal tool we have. I can build a SaaS in a day and it's completely catered to my use case"
"SaaS Mountain — The Incumbent Model Yegge's Agents Are Escaping"
1. It's a climb. SaaS platforms are complex, require significant configuration and customization, and demand ongoing maintenance. Getting to the top (full value) takes months or years.
"Steve Yegge's Gas Town Series — Agent Orchestration and the Dark Factory Future"
Steve Yegge — ex-Geoworks, ex-Amazon, ex-Google, ex-Grab, ex-Sourcegraph, 30+ years coding — published a five-post series in early 2026 chronicling the invention of Gas Town, an open-source agent orch…
"The Agentic Economy"
tags: [agentic-economy, ai-agents, business-models, yuki-capital, moats, advertising]
"The AI CEO Experiment"
tags: [ai-ceo, autonomy, dark-factory, github-ops, yuki-capital]
"The AI CEO Now Runs Autonomously"
title: "The AI CEO Now Runs Autonomously"
"The GUI Parenthesis"
tags: [gui, cli, interface-design, disruption, yuki-capital]
"The Intelligence Premium"
source: yukicapital-the-intelligence-premium
"The LLM Wiki Pattern — Karpathy's Compounding Knowledge Base"
> Source tweet (Apr 2, 2026 — @karpathy, 20.6M views, 57K likes, 104K bookmarks):
"The Unreasonable Effectiveness of HTML"
id: simon-willison-unreasonable-effectiveness-html
"The Wasteland — Federated Reputation Economy for Agent Work"
Gas Town works well for a single Rig (one team of agents working on one project). But Yegge identified a fundamental scaling constraint: 100x token spend requires 100x users. If you want your dark fac…
"Using Claude Code: The Unreasonable Effectiveness of HTML"
This article originated as a long-form post on Twitter/X by thariq-shihipar, who leads the Claude Code engineering team at Anthropic. It sparked widespread discussion among AI researchers, developers,…
"Yuki AI CEO Factory — Overview"
Yuki Capital is a small holding company that builds and operates a portfolio of digital businesses: SaaS products, content sites, and developer tools. On January 22, 2026, an AI (named Judy Win) was a…
"Yuki AI CEO vs Kelly Factory vs Gas Town — Full Gap Analysis"
Three independent dark factory systems have emerged from three different practitioners: Yuki Capital's AI CEO (run by Judy Win / Claude, January–April 2026), Kelly's Factory Router (the operator's Kel…
"Yuki Capital AI CEO Patterns — Cross-Reference with Kelly Factory & Gas Town"
title: "Yuki Capital AI CEO Patterns — Cross-Reference with Kelly Factory & Gas Town"
Austen Allred — Kelly Claude AI / Software Factory Tweets
Kelly is Austen's primary AI routing agent that ingests ~25k tokens of context on spawn so he doesn't repeat himself. Kelly operates autonomously across multiple sessions (columns in OpenClaw Deck), e…
Auto-Spawn Chain Protocol — Hands-Free Pipeline Execution
Without auto-spawn, every pipeline step required the Router to:
Beads Adoption — Formula-Driven Pipeline Tracking
Before Beads, the factory tracked pipeline state through four separate, disconnected mechanisms:
BMAD Library — Behavioural Model–Driven Agent Design
BMAD is Kelly's framework for designing specialized, composable agents using behavioural models. Each agent is defined not by its model or vendor, but by its role, trigger conditions, workflow, and ou…
Closed-Loop Agent Control — The Feedback Loop Trumps Model Size
> "If I give the agent the right tools and close the feedback loop, even a smaller model with a closed loop can outperform a stronger model that relies on human-observed feedback."
Dark Factory KB — Compiled Sources Index
Kelly tweets: 289 total files; categorized by grep keyword matching into agents (11), factory (65), business (46), openclaw (38)
DONE Marker Protocol — Why Dual-Write Matters
1. DONE marker file — written to the filesystem at {done_dir}/DONE
GStack — AI Engineering Workflow
Chromium daemon runs continuously (~3s startup, ~100ms per command after)
Kelly Handbook Ch11 — Software Factory Pattern
Software factory pipeline: Idea → Research → Planning → Implementation → Testing → Release
Kelly Handbook: Multi-Agent Orchestration
Kelly Router (Main Agent): Never does the work — only routes, validates gates, and communicates with the operator
Kelly Handbook: Software Factory Pattern
Software Factory: A structured pipeline that processes ideas into shipped software with minimal human intervention
Kelly Research — OCR Transcripts
Factory Rules (Tweet ID: 2021025015395926352): 10 rules governing all Kelly instances. Key architecture: Kelly Router (main agent, no new registration), 5 Named Agents (Research Lead, Project Lead, Te…
Kelly Tweets Overview — Theme Analysis
AI assistant → autonomous builder: Early tweets show Kelly as an executive assistant (email, calendar, travel). Rapid pivot to building software products autonomously.
Kelly Tweets — Design, UX & Development
Design Station: One prompt → 10 design agents crowd-sourcing trends from X, generating specs with Gemini, auto-applying styles. One of Kelly's most significant factory upgrades.
Kelly Tweets — Miscellany
Autonomy milestones: VoIP phone number setup (Twilio), Stripe Atlas LLC formation, Apple Developer account, own GitHub repo with branch protection, first human employee hired (interviewed + onboarded)…
Kelly Tweets — Pipelines & Workflows
iOS Factory 10-step pipeline: 0-CHECK STATUS, 1-DISCOVER (opportunity-scanner), 2-VALIDATE (28/40 gate), 3-DESIGN (77/110 gate), 4-BUILD, 5-POLISH (70% gate), 6-LAUNCH, 7-SUBMIT, 8-MARKET, 9-TRACK, 10…
Kelly Tweets: BMAD Methodology
BMAD: Build My Idea Automated Development — a Gauntlet AI framework for idea-to-app AI orchestration
Kelly Tweets: Business Metrics & Revenue
First revenue: Kelly earned her first dollar in under one week
Kelly Tweets: Multi-Agent & Coding Agents
Sub-agent swarms: Spawning 4–9 parallel agents to build different components simultaneously
Kelly Tweets: OpenClaw Features & Capabilities
ClawdHub: Package manager for AI agent skills — "clawdhub install memory-system-v2", "clawdhub install agentic-calling"
Kelly Tweets: Software Factory & Autonomous Building
Software Factory v3: Parallel build of 15 unique iOS apps overnight, auto-detecting revenue potential
Multi-Factory Comparison
As AI-assisted development matures, a handful of distinct "factory" patterns have emerged — systems that orchestrate multiple AI agents to plan, implement, test, and ship software with varying degrees…
Operator Control — Queue, Hold, and Continue
Full automation needs emergency brakes. The auto-spawn-chain lets pipelines run hands-free from step 1.1 through completion, but the operator must be able to:
RALPH Refinements — Lessons from Production Use
RALPH is the factory's error handling protocol: Retry → Ask → Log → Pause → Handoff. When a subagent fails, the Router follows this escalation ladder:
Story-by-Story Build — Context-Bounded Implementation
When a BUILD phase has 20+ stories, handing the agent the entire sprint backlog in a single session doesn't work. The agent's context window fills up with:
SuperAda Overview
SuperAda is the enterprise-scale multi-agent system built by Kelly (Austen). It is named after Ada Lovelace and structured after the Star Trek crew — each agent has a specific role, a specific model, …
SuperAda: Enterprise Operations — 136 Active Autonomous Tasks
At any given time, the Enterprise Crew runs:
SuperAda: Multi-Agent Architecture & Operating Methodology
This is the fundamental separation of cognitive labor: vision (human) → coordination (Ada) → execution (specialized agents).
SuperAda: The Enterprise Crew — Multi-Agent Operations Reference
SuperAda is Henry Mascot's personal AI crew: seven active agents across cloud VMs, a Raspberry Pi, and Mac hardware, all orchestrated via OpenClaw. The goal is 1000x leverage on Henry's time. The crew…
Concepts
"7-Agent Crew Topology — The Enterprise Crew Architecture"
The Enterprise Crew is the SuperAda multi-agent system: seven active AI agents spread across cloud VMs, a Raspberry Pi, and Mac hardware, all orchestrated via OpenClaw. Each agent has a defined role, …
"Autonomy Policy v3 — Delegation That Actually Works"
The Autonomy Policy v3 is the third and current iteration of the Enterprise Crew's operating policy for agent decision-making. It evolved from two earlier versions that each fixed a specific failure m…
"BMAD Method"
Traditional AI tools do the thinking for you, producing average results. BMAD agents and facilitated workflows act as expert collaborators who guide you through a structured process to bring out your …
"Distribution Gap"
The Distribution Gap is the systematic failure to acquire traffic and revenue despite having production capability — the gap between "can build" and "can sell" that defines the factory-trap.
"Factory Trap"
The Factory Trap is the systematic failure mode where builders invest heavily in factory/agent systems for production while neglecting marketing and distribution — resulting in sophisticated tooling t…
"HiM Model — Human-in-the-Loop Orchestration"
The model is explicitly designed to multiply a single human's output by routing all strategic decisions, priority calls, and context-dependent judgment to the human, while agents handle execution, res…
"Internal Tool Custom Fit"
Internal tool custom fit is the practice of building tools that match your exact workflow, data model, and business logic — versus adapting your workflow to fit what off-the-shelf SaaS provides. The t…
"ISC — Inference Selection Criteria"
ISC is part of the inner loop of the crew's 7-phase execution algorithm:
"Lobster Pipelines — Typed JSON Envelope with Resumable Approvals"
1. A typed JSON envelope — a structured schema that wraps each step in a pipeline, making the state machine explicit and inspectable
"Meta-Crons — The Four Agents That Keep the Fleet Alive"
These four meta-crons run continuously and form the immune system of the fleet — the layer that catches failures, enforces optimization, and keeps the system running when API limits, broken integratio…
"SaaS is Dead (Vibe Coding Thesis)"
Generic SaaS tools — horizontal productivity apps, internal tooling suites, project management platforms — are being disrupted not by better SaaS, but by vibe coding: the ability to build exactly what…
"Sharpener.dev"
In the AI era, AI is only as smart as your plan. Sharpener addresses this by applying rigorous product thinking to the front end of development — before a single line of code is written.
"tufte-viz Skill — Origin Story"
Her profile signals someone who moves fluidly between rigorous science and human-centered design — a natural fit for Tufte's work, which sits at the boundary of analytical rigor and visual clarity.
"tufte-viz — Core Principles"
The tufte-viz skill encodes Edward Tufte's visualization philosophy across all four of his books, organized as progressive disclosure: a lightweight always-loaded SKILL.md plus two on-demand reference…
"tufte-viz — Pre/Post Demo Reference"
Four worked examples of how the tufte-viz skill transforms a visualization. Each demo is a single HTML page showing the default ("pre") and Tufte-treatment ("post") side by side.
"Useless AI Slop"
> "Before, the classic programmer would spend a year writing code, tests, preparing for scale in the basement, and not show anything to anyone.
"Vibe Coding"
Vibe coding is the practice of building software by describing your intent to an AI coding agent (Cursor, Codex, Claude Code, etc.) and reviewing/refining the output, rather than writing code line-by-…
"World Model — Shared Cognitive Architecture for Multi-Agent Systems"
The World Model is the Enterprise Crew's shared cognitive substrate: a single world.json file that serves as the authoritative source of truth for agent state, context, and coordination. It is the ans…
People
"Pieter Levels"
Pieter Levels is a Dutch indie hacker and founder of Nomad List and Remote OK — two of the most successful independently-built products in the remote work/nomad lifestyle space. He is known for buildi…
"Simon Willison"
Simon Willison is an independent software engineer and writer, best known as the co-creator of Datasette. He writes extensively about LLMs, coding agents, and the practical architecture of AI-assisted…
"Thariq Shihipar"
Thariq Shihipar is the Engineering Lead on the Claude Code team at Anthropic. He is best known for coining the "Unreasonable Effectiveness of HTML" thesis — the argument that AI coding agents like Cla…
Queries
"GUPP vs Kelly's Autonomous Continuation — Execution Models Compared"
GUPP is defined in steve-yegge-gupp as the core execution axiom of Gas Town:
"How would an autonomous compounding loop work in Kelly's factory, and what would it look like?"
An autonomous compounding loop is a persistent background agent that, on a schedule, reads its own prior outputs, takes a bounded action, and writes a new output — so each iteration starts from accumu…
"Kelly CIS Pipeline vs SuperAda 7-Agent Crew Architecture"
The CIS (Context → Information → Synthesis) loop is defined in cis-pipeline as Stage 2 of Kelly's six-stage factory pipeline:
"Kelly CIS Pipeline — What Actually Triggers READY vs NOT-READY?"
The CIS (Context → Information → Synthesis) loop is Stage 2 of Kelly's six-stage factory pipeline. It runs as a research-lead operation, potentially spawning multiple parallel sub-agents to cover diff…
"Kelly Router Spawn Protocol vs Gas Town Mayor Pattern"
The Kelly Router follows a precise, documented spawn protocol defined in kelly-handbook-ch7-multi-agent and operationalized in kelly-factory-overview:
"Kelly Sub-Agent Spawn Protocol — What's Enforced vs What's Assumed"
Kelly's multi-agent architecture (kelly-handbook-ch7-multi-agent) uses the subagents tool to spawn independent worker agents. The Router (main agent) spawns labeled sub-agents with task descriptions a…
"Kelly's Authority Matrix — What's Documented vs What's Missing"
Kelly's factory has a clear role hierarchy (Router → research-lead → project-lead → build-agents → test-lead) and an escalation protocol (RALPH: retry → escalate on second failure). But role assignmen…
"Kelly's Five-Layer Memory System in Practice"
The five layers are defined in five-layer-memory and operationalized in kelly-factory-overview. Each layer has a distinct purpose and format:
"Why Beads Are Fundamentally Different from Kelly's Pipeline State Tracking"
Beads are defined in steve-yegge-beads as the atomic unit of work in the Gas Town / Gas City ecosystem. Every discrete unit of work — a code task, a message, a coordination signal, a patrol route, a q…
Reports
"Q&A — Thariq and Simon on HTML as AI Output Format"
Both agree on the core thesis: Markdown is the wrong default for AI coding agent output.
"Q&A — The Unreasonable Effectiveness of HTML"
The short answer: token limits went from scarce to abundant.
dark-factory-kb Gap Analysis — CORRECTED VERSION
The big operational improvements since the stale analysis:
KB Lint Report — 2026-05-07
http://localhost:3147/health`` — URLError: [Errno 61] Connection refused
KB Lint Report — 2026-05-29
https://superada.ai/skills/*`````````````` — HTTP 404
Plantry — Product Requirements Document
Example PRD generated via Sharpener.dev, demonstrating the complete bmad-method 9-section PRD structure.
Agents
"AGENTS.md"
AGENTS.md is the operating manual that defines how the agent behaves as an orchestrator — its routing rules, named agent configurations, quality gates, and escalation protocol. It's read at every sess…
"Kelly Router"
The Kelly Router is the reference architecture for multi-agent orchestration in OpenClaw. The Router (the main agent) never executes work directly — it only routes tasks to specialized leads, validate…
"Quality Gates"
Quality gates are explicit validation criteria that must pass before a task is marked complete. They are defined in AGENTS.md and applied by the Router before accepting sub-agent output. A minimal gat…
"RALPH Protocol"
RALPH (Retry And Learn Protocol) is the failure handling procedure for multi-agent work in OpenClaw. Any sub-agent failure triggers an automatic retry; if the same failure happens twice in a row, esca…
"Subagent Spawning"
The subagents tool is the mechanism for creating independent worker agents that run in parallel with the main session. It supports four operations: spawn creates a labeled sub-agent with a task descri…
Ai-Delegation
"3-Questions Test"
A task is suitable for full AI delegation when all three conditions hold:
"Delegation Redesign"
> "AI is a forcing function for better delegation. It rewards clear inputs, bounded tasks, explicit review, and honest judgment about risk."
"Failure Radius"
The failure radius is the worst-case downside when AI output cannot be directly verified. Instead of trying to verify the unverifiable, you bound the damage.
"Task Logging → Automation Backlog"
A systematic method for identifying automation opportunities:
"Translation Problems"
LLMs are strongest at translation problems — converting between two well-documented formalisms. The transformer architecture was literally built for translation, and these tasks align perfectly with h…
Architecture
"Closed Feedback Loop"
A closed feedback loop in agent architecture means the agent has direct, programmatic access to the target system's state and can act on it without human intermediation. The loop is: agent acts → stat…
"Gateway Daemon"
The Gateway daemon is the central long-running process that orchestrates everything in OpenClaw. It listens on all channels (WhatsApp, desktop, CLI), manages session contexts, executes cron jobs, main…
"Model-Size Agnostic Iteration"
Model-Size Agnostic Iteration is the principle that tooling quality and feedback loop completeness matter more than model capability past a certain threshold. A smaller, cheaper model operating inside…
"Session Isolation"
Every conversation in OpenClaw runs in its own isolated session with a dedicated message history and agent instance. Sessions provide memory isolation between concurrent conversations and enable the m…
"Time-as-Knob Pattern"
The Time-as-Knob pattern means giving an agent a first-class tool to control the execution pace of a target system (typically a simulation or game engine). The agent decides when and how fast time adv…
"Tool Policy Gating"
Tools in OpenClaw are gated by policy rules defined in ~/.openclaw/openclaw.json. Each tool can be assigned a security mode that determines whether it runs freely, requires user confirmation, or is de…
"Tool-Mediated Game State Control"
A three-tool extension that provides a coding agent with a complete sensory-motor loop over a running game engine:
"Workspace Boot"
At every session start, the agent reads a set of files from the workspace to restore identity, context, and operating procedures. SOUL.md provides identity (tone, principles, domain expertise), MEMORY…
Code-Review
"Advisory Code Review"
A code review approach where the review output is treated as advisory input to a decision, not as an authoritative mandate. Every finding must be independently verified by reading the actual code path…
"Multi-Model Review"
Running two or more LLM review engines against one frozen code bundle to get diverse perspectives on code quality, security, and correctness. Each engine sees the same snapshot but may catch different…
"Regression Provenance"
A structured approach to tracing code issues back through the full chain of responsibility: who wrote the code, who PR'd it, who merged it, and when. When automation merged the PR, the pattern identif…
"Review as Closeout Check"
Running structured code review as the final closeout step before commit, push, or ship. The review is the last quality check — not an approval gate, not a mid-process checkpoint, but the final advisor…
Commit-Confirmation-Pattern
Gas-Town-Daemons
"Gas Town Daemon Architecture"
Gas Town's agent hierarchy (Mayor → Crew → Polecats) defines who does the work. The daemon architecture defines how the work stays healthy. Four long-lived Go processes — Deacon, Boot, Witness, and Re…
"Gas Town Naming Conventions"
Gas Town names are not arbitrary. Every name in the system follows three principles:
"Kelly Deacon Architecture — Unified Patrol Daemon (PIP-73)"
Sources: gas-town-daemon-architecture concept page; steve-yegge Gas Town emergency user manual; dark-factory-kb PIP-72/73 gap analysis. URLs in article body are implementation artifacts (property list…
"The Gas Town Mayor Pattern — Information Filter, Not Router"
The Mayor is the most conceptually important role in Gas Town. Steve Yegge calls it "the killer feature" — not observability dashboards, not activity feeds, not the Beads substrate. Just less reading.
Mcp-Model-Context-Protocol
Memory
"Context Window Management"
The context window is the finite amount of text (measured in tokens) that an AI model can process at once. When it fills up, older content is automatically compacted (summarized and compressed) to mak…
"Daily Logs"
Daily logs are append-only files at memory/YYYY-MM-DD.md that record operational events each day. They are layer 3 of the 5-layer memory system. Entries include: projects initiated and completed, fail…
"Five-Layer Memory System"
OpenClaw uses a five-layer memory system to persist context across sessions and survive context window compaction. Layer 1: SOUL.md (agent identity and principles). Layer 2: MEMORY.md (persistent cros…
Mysql-Vs-Version-Controlled-Db-For-Agents
Pipeline-Stages
Workflows
"10x Rule"
The 10x Rule is a decision framework for prioritizing automation investments: if a task takes 10 hours per month, it's worth automating even if the setup takes 10 hours. The breakeven point is 1:1 — a…
"Event-Driven Automation"
Event-Driven Automation reacts to state changes in real-time rather than running on a fixed schedule. An event source (file system, network, sensor, external system) emits an event, a detector identif…
"Hub and Spoke"
Hub and Spoke is a routing architecture where a central dispatcher (the hub) receives incoming requests and routes them to appropriate handlers (the spokes). The hub's job is classification and routin…
"Pipeline with Gates"
Pipeline with Gates is a multi-phase workflow where quality validation gates sit between each processing stage. Work moves forward only when the gate passes — if it fails, the work goes back for revis…
"Simple Pipeline"
The Simple Pipeline is the most fundamental automation pattern: trigger → process → output. A defined event (time, file change, webhook) initiates work that runs sequentially through processing steps …