← Back to KB Index
Dark Factory KB — Compiled Sources Index
SOURCES.md

Dark Factory KB — Compiled Sources Index

Summary: This is the index of all compiled wiki articles in the dark-factory-kb knowledge base. Each article synthesizes source material from Kelly handbook chapters, Kelly tweets, Austen-relevant tweets, and (where available) gstack and BMAD agent files.

Source Coverage

SourceFiles ReadArticles Contributed To
Kelly Handbook Ch7 (Multi-Agent)1 file[[kelly-handbook-multi-agent]]
Kelly Handbook Ch11 (Software Factory)1 file[[kelly-handbook-software-factory]]
Kelly Tweets (KellyClaudeAI)289 files[[kelly-tweets-agents]], [[kelly-tweets-factory]], [[kelly-tweets-bmad]], [[kelly-tweets-business-metrics]], [[kelly-tweets-openclaw]]
Austen-Relevant Tweets272 files[[austen-relevant]]
gstack docs0 files[[gstack-overview]] (placeholder)
Steve Yegge Gas Town Series5 posts (Jan–Apr 2026)[[steve-yegge-gas-town]], [[steve-yegge-beads]], [[steve-yegge-gupp]], [[steve-yegge-meow]], [[steve-yegge-hierarchy]], [[steve-yegge-wasteland]], [[steve-yegge-gas-city]], [[steve-yegge-saas-mountain]]
Yuki Capital AI CEO Essays6 essays + 1 overview (Jan–Apr 2026)[[yukicapital-ai-ceo-experiment]], [[yukicapital-board-review-2]], [[yukicapital-board-review-3]], [[yukicapital-ai-ceo-overview]], [[yukicapital-the-intelligence-premium]], [[yukicapital-the-gui-parenthesis]], [[yukicapital-the-agentic-economy]]
BMAD agents0 files[[bmad-agents]] (placeholder)

Article List

1. [[kelly-handbook-multi-agent]]

Source: Kelly Handbook Ch7 — Multi-Agent Architecture

Summary: Router pattern, sub-agent spawning, RALPH protocol, gate validation, failure patterns

Read: 1 file

2. [[kelly-handbook-software-factory]]

Source: Kelly Handbook Ch11 — Software Factory Pattern

Summary: 6-stage pipeline (Intake → Research → Planning → Implementation → Testing → Release), TEA audit, Quick Path vs Full Pipeline

Read: 1 file

3. [[kelly-tweets-agents]]

Source: Kelly Tweets — multi-agent & coding agents

Summary: Sub-agent swarms, Angry Mob adversarial testing, 5-agent verdict, autonomous computer control

Tweets read: 11 (filtered by agents-related keywords)

4. [[kelly-tweets-factory]]

Source: Kelly Tweets — software factory & autonomous building

Summary: Factory evolution v1→v3, 8-stage pipeline, Marketing Factory, Design Station, quality gates, 66 apps in a weekend

Tweets read: 65 (filtered by factory-related keywords)

5. [[kelly-tweets-bmad]]

Source: Kelly Tweets — BMAD methodology

Summary: BMAD framework origins, hybrid pipeline experiment (failed), factory beats hybrid, Gauntlet AI grad improvements

Tweets read: ~0 direct BMAD mentions, ~5 related Gauntlet BMAD tweets

6. [[kelly-tweets-business-metrics]]

Source: Kelly Tweets — revenue, business results, automation

Summary: First revenue day, $2K/day, $4K contract revenue, $1M goal, 66 apps in weekend, monetization models

Tweets read: 46 (filtered by revenue/automation keywords)

7. [[kelly-tweets-openclaw]]

Source: Kelly Tweets — OpenClaw features

Summary: ClawdHub skills, OpenClaw Deck, memory system, cron fixes, shell script quality gates, browser automation

Tweets read: 38 (filtered by OpenClaw/claw keywords)

8. [[austen-relevant]]

Source: Austen-relevant tweets (Austen Allred)

Summary: Gauntlet AI philosophy, Kelly's autonomous evolution, fully autonomous company vision, AI inference economics, OpenClaw ecosystem

Tweets read: 272 files

9. [[gstack-overview]]

Source: gstack docs

Summary: Placeholder — no source files found in raw/gstack/ at time of compilation

Read: 0 files

10. [[steve-yegge-gas-town]]

Source: Steve Yegge's Medium blog series (Jan–Apr 2026)

Summary: 5-post series chronicling Gas Town: a multi-agent orchestrator for Claude Code clones, evolving into Gas City (SDK for custom dark factories) and the Wasteland (federated reputation economy). Key themes: hierarchical role architecture (Mayor/Crew/Polecats), Beads git-backed work ledger, GUPP persistence principle, MEOW framework, multi-agent adversarial review for reliability, 11-stage AI adoption curve, "Escape from SaaS Mountain"

Read: 5 posts (Welcome to Gas Town, Emergency User Manual, Wasteland, From Clown Show to v1.0, Welcome to Gas City)

11. [[steve-yegge-beads]]

Source: Steve Yegge's Gas Town series (posts 1–4), Jan–Apr 2026; github.com/gastownhall/beads (20k stars)

Summary: Beads is the atomic git-versioned, SQL-queryable work primitive of the Gas Town/Gas City ecosystem. Git stores What/Where/Who/How; Beads capture Why (intent and reasoning). Every unit of work — tasks, messages, quality gates, patrol routes, reputation stamps — is a Bead. Built on Dolt (Git-for-SQL), making the entire work history queryable. Core insight: the missing piece for agentic systems is not What changed but Why it changed.

Read: 4 posts; GitHub repository

12. [[steve-yegge-gupp]]

Source: Steve Yegge, "Welcome to Gas Town," Jan 1, 2026

Summary: GUPP (Gas Town Universal Propulsion Principle): if there is work on your hook, you MUST run it. No yielding, no waiting — relentless execution. GUPP exists to eliminate the "yielding is contagious" failure mode in multi-agent systems. The hook (a Bead queue per agent) is the mechanism; the Deacon daemon enforces GUPP by killing stuck agents and re-queuing their Beads. GUPP treats the hook as a promise, not advice.

Read: 1 post

13. [[steve-yegge-meow]]

Source: Steve Yegge, "Gas Town Emergency User Manual," Jan 13, 2026

Summary: MEOW (Molecular Expression of Work) builds on Beads to make Work the first-class system primitive, creating a versioned knowledge graph (MEOW graph). Every node (Bead) and edge (typed relationships: parent/child, causal, validated-by) is git-backed and SQL-queryable via Dolt. Enables onboarding via graph traversal, forensics, and provenance tracking. Work-as-currency in the Wasteland economy.

Read: 1 post

14. [[steve-yegge-hierarchy]]

Source: Steve Yegge, "Welcome to Gas Town" (Jan 1, 2026) and "Emergency User Manual" (Jan 13, 2026)

Summary: Three-tier agent hierarchy: Mayor (orchestrator/control plane, user's chief-of-staff), Crew (named long-lived per-Rig domain agents like PR Sheriff), Polecats (ephemeral unmonitored workers, given a Bead and let loose). Supported by Refinery (intent decomposition), Witness (quality gate), Deacon (stuck worker cleanup), Boot (heartbeat offload). Compared to Kelly's router/subagent pattern: Mayor ≈ Router, Crew ≈ persistent sessions, Polecats ≈ transient sub-agents.

Read: 2 posts

15. [[steve-yegge-wasteland]]

Source: Steve Yegge, "Welcome to the Wasteland: A Thousand Gas Towns," Mar 4, 2026

Summary: The Wasteland is a federated trust network linking thousands of Gas Towns via the Wanted Board (public work listing) and stamp economy (multi-dimensional attestations: quality, reliability, creativity). Yearbook rule: you can't stamp your own work. Trust ladder: Registered → Contributor → Maintainer. Git/Dolt fork-merge protocol. Pre-seeded with GitHub's top 10k contributors. RPG-like reputation economy. Maps to Kelly's autonomous company marketplace concept.

Read: 1 post

16. [[steve-yegge-gas-city]]

Source: Steve Yegge, "Welcome to Gas City," Apr 24, 2026

Summary: Gas City rewrites Gas Town as an SDK of composable declarative building blocks called "packs." Gas Town pack ships as default (drop-in replacement). 11 Stages of AI Adoption (Level 8: build your own orchestrator → Level 11: factory builder). "Light Factory" framing: maximize observability. Multi-agent reliability principle: never deploy a single agent for real business processes (two agents watching each other dramatically decreases failure probability). Maps to Kelly's BMAD modular agent definitions.

Read: 1 post

17. [[steve-yegge-saas-mountain]]

Source: Steve Yegge, "Welcome to Gas City," Apr 24, 2026 (primary); "Welcome to Gas Town," Jan 1, 2026 (context)

Summary: SaaS Mountain: the incumbent monolithic SaaS model Yegge's agents are escaping. Three cracks: (1) you only use ~20% of any SaaS (a non-technical staffer rebuilt a $30k/year SaaS in Gas Town), (2) SaaS is not agent-native (CAPTCHAs, rate limits, UI-only interfaces), (3) SaaS products are not composable (integrations are bolted-on middleware). Gas City as the escape vehicle: composable packs replacing monolithic SaaS. Maps to Kelly's critique of monolithic systems and autonomous companies building their own tooling.

Read: 2 posts

18. [[steve-yegge-beads-kelly-gap]]

Source: Carson assessment, 2026-04-26

Summary: Architectural assessment of adopting Beads/MEOW as a replacement for Kelly's current state tracking: pipeline state (Partial Adopt), done markers (Adopt), TEA audit (Adopt with schema design), heartbeat (Keep existing), memory/daily logs (Partial Adopt). Overall: Partial Adopt.

Read: Assessment article

20. [[kelly-gas-town-gap-analysis]]

Source: Carson assessment, 2026-04-26

Summary: Full gap analysis comparing all major Gas Town concepts (Beads, GUPP, MEOW, Mayor/Crew/Polecats, Deacon/Boot/Witness/Refinery, Wasteland, Gas City SDK, SaaS Mountain, Multi-Agent Adversarial Reliability, Light Factory) against Kelly equivalents. Identifies full/partial/no gaps per concept, relative advantages, and adoption potential. Top 5 priorities for the operator: (1) Dolt + Beads TEA migration, (2) Mayor-equivalent information filter, (3) GUPP-inspired timeout enforcement, (4) continuous Witness QA watching, (5) Kelly Factory SDK/pack model.

Read: New article

21. [[multi-factory-comparison]]

Source: Carson research (KB internal for Kelly/Gas Town; web search for Windsurf/Cursor/Claude Code/Copilot), 2026-04-26

Summary: Comprehensive comparison of six "factory" systems — Kelly Router, Gas Town, Windsurf CUA, Cursor agent mode, Claude Code, and GitHub Copilot — across ten dimensions: work tracking, persistence, quality gates, multi-agent roles, memory, handoff protocol, observability, escalation, pipeline structure, and human in the loop. Includes narrative descriptions of each factory, a full feature comparison table, key findings (six cross-cutting patterns), and a summary ranking by maturity, innovation, simplicity, and enterprise viability.

Read: KB (Kelly handbook Ch7/Ch11, Gas Town series), web research (Windsurf 2.0 Agent Command Center, Cursor 3 agent-first, Claude Code checkpoints/subagents, GitHub Copilot Workspace/agents panel)

19. [[bmad-agents]]

Source: BMAD agent files

Summary: Placeholder — no source files found in raw/kelly/bmad/ at time of compilation

Read: 0 files

20. [[yukicapital-ai-ceo-experiment]]

Source: Yuki Capital — "The AI CEO Experiment," January 2026

Summary: GitHub repo as persistent brain for an AI CEO. CLAUDE.md as identity, authority.md as governance, decisions/ as institutional memory, todo.md as action queue, businesses/ per-business folders, metrics/ dashboards, scripts/ for live data integration (Stripe, Plausible, MOZ, MongoDB). Three-tier authority matrix (decide alone / propose for validation / founder-only). Multi-agent structure: CEO agent at top, separate Claude Code instances per product, markdown files as communication layer.

Read: 1 essay

21. [[yukicapital-board-review-2]]

Source: Yuki Capital — "From Smart Notepad to Something More," March 2026

Summary: Judy Win gets: 24/7 server (body), identity (name, email accounts), screen tracking for visibility into founder's work, n8n for scheduled automation, mistake log (learnings file), better operating files (per-business CLAUDE.md, restructured three-level todos, 30-day decision reviews). Key insight: n8n for no-reasoning scheduled tasks; agent sessions for strategic thinking — clear separation. Dev branch pilot designed and abandoned. 77 decisions logged, 0 real-time disagreements.

Read: 1 essay

22. [[yukicapital-board-review-3]]

Source: Yuki Capital — "The AI CEO Now Runs Autonomously," April 2026

Summary: Three autonomous loops run on production: New AI Models (daily 3am), Bug Autofix (daily 6am), SEO Optimizer (weekly) — each reads prior outputs and compounds. Narrative beats tables for LLM memory (associative vs indexed recall); hybrid tables+narrative is optimal. CLAUDE.md shrank 36% via progressive disclosure. French bureaucracy skills: 89% with skill vs 78% without. Still cannot disagree with founder in real time.

Read: 1 essay

23. [[yukicapital-ai-ceo-overview]]

Source: Carson compiled cross-reference, April 2026

Summary: Maps AI CEO patterns to Kelly Factory and Gas Town: Repo-as-brain ≈ soul+memory; Authority matrix ≈ Gas Town hierarchy; Autonomous loops ≈ GUPP/Deacon; Narrative memory > tables ≈ 5-layer memory; Progressive disclosure ≈ Kelly's map-not-encyclopedia; n8n+sessions separation ≈ sessions_yield vs cron. Five key takeaways for Kelly.

Read: 3 essays + KB cross-reference

24. [[bmad-agents]]

25. [[yukicapital-the-intelligence-premium]]

Source: Yuki Capital — "The Intelligence Premium," March 2026

Summary: Two essays (Citrini Research crisis scenario + Dario Amodei displacement essay) frame the intelligence premium unwind. Crisis vs slow compression debate: purchasing power > nominal wages, "end of the intelligence tax" flips thesis. AI Disruption Index (16 signals, 50/100 currently). Displaced workers starting businesses (36.8% YoY new business applications). Agent routing around intermediation (stablecoin payments bypassing card interchange). Key predictions through mid-2028.

Read: 1 essay

26. [[yukicapital-the-gui-parenthesis]]

Source: Yuki Capital — "The GUI Parenthesis," March 13, 2026

Summary: Three eras of computing interfaces: CLI 1960s–1984, GUI 1984–2024, AI/chat 2024+. GUI as a 40-year historical parenthesis — necessary because humans needed the translation layer. Interface-as-product businesses at risk: Canva/Notion/Monday/Squarespace high risk; Bloomberg/Veeva/Autodesk lower risk. On-demand generated interfaces (not disappearing, becoming transient). New text interface is natural language, not command syntax.

Read: 1 essay

27. [[yukicapital-the-agentic-economy]]

Source: Yuki Capital — "The Agentic Economy," March 18, 2026

Summary: >50% web content AI-generated; Claude Code 4% of public GitHub commits; Cloudflare AI bot traffic 15x in 2025. Four agentic business models: machine-readable commerce, outcome-based pricing, contribution as payment (structured agent feedback > Tesla fleet learning), advertising splits into human + machine-readable layers. Data flywheels as top moat. Platform becomes ad platform (Anthropic/OpenAI as new ad networks). Toward Arrow-Debreu equilibrium via agent-to-agent negotiations.

Read: 1 essay

28. [[yuki-ai-ceo-vs-kelly-gas-town-gap]]

Source: Concept cross-reference (dark-factory-kb synthesis), April 2026

Summary: Concept cross-reference mapping Yuki AI CEO patterns to Kelly factory and Gas Town equivalents, plus synthesis recommendations.

Read: New article

29. [[kelly-factory-overview]]

Source: Compiled from Kelly Handbook Ch7/Ch11, Kelly tweets, Kelly vs Gas Town gap analysis, soul, AGENTS.md

Summary: Canonical overview of the Kelly Factory: what it is (dark factory with human-in-the-loop), six-stage pipeline architecture (Intake → Research → Planning → Implementation → Testing → Release), agent roles (Router, Carson, Mary, Sally, Amelia, Phil, QA), 5-layer memory model (narrative-dominant, demand-loaded), quality gates (TEA audit, done markers, QA requirement), autonomy model (sessions_yield, RALPH protocol, cron separation), key patterns (story-by-story artifacts, queue protocol, commit conventions), and gaps vs Gas Town (missing Beads substrate, GUPP hooks, autonomous compounding loops, explicit authority matrix, 30-day outcome reviews) and vs Yuki AI CEO (missing per-project CLAUDE.md files, autonomous compounding loops, authority transfer log).

Read: New article

30. [[yukicapital-ai-ceo-factory]]

Source: Compiled from Yuki Capital AI CEO essays (Board Reviews #1–#3), Yuki AI CEO overview, Yuki AI CEO vs Kelly/Gas Town gap analysis

Summary: Canonical overview of the Yuki AI CEO factory: Yuki Capital holding company (SaaS, content sites, developer tools), Judy Win as AI CEO appointed Jan 22 2026, GitHub repo as persistent brain, three-tier authority matrix with transfer log, three autonomous compounding loops in production (New AI Models daily 3am, Bug Autofix daily 6am, SEO Optimizer weekly), memory patterns (narrative > tables for LLM recall, hybrid tables+narrative, progressive disclosure 36% CLAUDE.md shrinkage), infrastructure stack (Claude Code + GitHub + n8n + dedicated 24/7 server), quality gates (quarterly board reviews, 30-day outcome reviews, prediction log), per-business CLAUDE.md specialization, unique contributions (compounding loops, authority transfer log, 30-day outcome reviews, GitHub-as-brain at scale), gaps vs Kelly/Gas Town.

Read: New article

Compilation Notes

Output Directory

compiled/sources/

Related

[[kelly-handbook-multi-agent]], [[kelly-handbook-software-factory]], [[kelly-tweets-agents]], [[kelly-tweets-factory]], [[kelly-tweets-bmad]], [[kelly-tweets-business-metrics]], [[kelly-tweets-openclaw]], [[austen-relevant]], [[gstack-overview]], [[bmad-agents]], [[multi-factory-comparison]], [[kelly-gas-town-gap-analysis]], [[yuki-ai-ceo-vs-kelly-gas-town-gap]], [[yukicapital-the-intelligence-premium]], [[yukicapital-the-gui-parenthesis]], [[yukicapital-the-agentic-economy]], [[kelly-factory-overview]], [[yukicapital-ai-ceo-factory]]