###/changelogwhat shipped
kevin@toronto:~/changelog $ git log --since="forever" --shipped
what shipped, when.
No marketing language. No "exciting announcements." Just dates and what actually changed — every post, every product page, every site update. Auto-generated from the post archive; curated highlights at the top when they exist.
###curated highlights
-
ship /changelog page goes live →Auto-generated chronological log of every post + curated site-update highlights. Operator-honest — no marketing language.
-
ship Related posts + share buttons + in-page search shippedFive UX improvements live: cache headers, RSS feed, /posts live search, X/LinkedIn/copy share buttons, category-matched related posts at bottom of every post.
-
ship RSS feed live at /feed.xml →48 items, auto-rebuilt on every Vercel deploy via scripts/build-feed.mjs.
-
seo Codex link + optimization audit shipped (round 1 + round 2)Two passes. Fixed 5 broken links, 4 missing product <h1>s, 5 orphaned published posts now in sitemap, skill count reconciled across 8 live files, Person + WebSite JSON-LD added, post-template fallback metadata, og:image on /now + /contact, og:url on /404.
-
ship Standalone /contact page →Self-contained operator-terminal page, ports the homepage send-a-ping form, fixes ~9 blog-post CTA 404s.
-
ship L3 free-fleet-research Phase 1competitor-scan for @KZic — Groq 70b extracts topics/hooks/CTAs/engagement signals from each tracked competitor's recent X timeline; optional Cloudflare Workers AI sentiment pass. $0 cost (free tiers). agents/free-fleet-research/.
-
ship Email revival flow: /stay + /api/revival-confirm →HMAC-token confirm landing for the 2026 list win-back. Pairs with scripts/scan-emails.mjs + agents/email-revival/merge-lists.mjs.
-
ship Brand-guard pre-commit hook + multi-brand voice enforcementSingle source of truth for voice rules; 4 wired gates (atomize-publisher, drift-check, push to Typefully, manual cli).
###may 202617 posts
-
Build Diary The Free Fleet Always-Busy Pattern: An Editor-in-Chief AI Manages a 7-Provider Content EngineSeven free LLM providers, $0 cost, idle 95% of the day — until we hired a robot editor.
-
Methodology Operationalize Every Fix: Turning 1 ESM Bug Into an 8-File Regression TestThe patch is the down payment. The sweep is the principal. Pay both.
-
Field Report Anatomy of an 87% Dispatch Failure: How One Missing Schema Field Killed Our Agent Fleet49% of our agent dispatches were failing for three weeks. The bug was one stale field in the adapter.
-
Methodology 18 Opus Workers, 4 Waves, $10: A Parallel-Agent Audit PatternHow to audit 39 agents and 100 markdown files in 3 hours for $10. The trick is non-overlapping domains.
-
Operator Manifesto Single Source of Truth: How 5 Fragmented Docs Became One Canonical ReferenceDoc rot is a tax on every reader. Pay it down once; collect interest forever.
-
Field Report 10 Real Agent Failures + How I Diagnosed ThemTen real OpenClaw fleet failures from my 36-agent stack, the signals that exposed each one, and the preventive fixes I shipped.
-
Field Report 4 Shell-Injection Vectors I Caught In My Own AgentsA security field report on four prompt-to-shell injection paths I found while hardening my OpenClaw agent fleet.
-
Field Report Config Drift Killed My Deployment — Here's the Guard I Builtopenclaw.json is 139KB of routing + agent config + model aliases. Unauthorized changes break things silently. Here's the SHA256 guard I built after one Tuesday morning broke everything.
-
Provider Economics Cost-Optimized LLM Routing: When to Use Claude, GPT, Grok, and Free ModelsDecision tree across 25+ LLM providers. Real cost data. Why I run 200+ models at $2.48/agent/day, and how the free-fleet optimizer routes 80% of work to $0/month.
-
Build Diary How My Agents Reflect OvernightA build diary on dreaming.sh, the overnight OpenClaw digest that turns memory, logs, and corrections into morning context.
-
Tools Every Script That Runs My FleetA compact index of the 30 OpenClaw automation scripts I use to keep my 36-agent fleet healthy, routed, checked, and improving.
-
Architecture How I Know All 36 Agents Are Alive Right NowThe file-based heartbeat layer I use to detect quiet agents, separate warnings from resets, and keep a 36-agent fleet observable.
-
Architecture How My 36-Agent Fleet Is Wired UpA diagram-heavy look at the router, dispatcher, workers, heartbeats, cost controls, and trust tiers behind my 36-agent OpenClaw fleet.
-
Build Diary How My Agents Find WorkHow auto-dispatcher.js turns Scout signals and unassigned Paperclip issues into routed work for my OpenClaw agent fleet.
-
Build Diary My Daily 8AM Briefing From The FleetHow morning-brief.js turns overnight fleet signals into one focused 8AM operator document with health, cost, and queue context.
-
Build Diary Scout: The Agent That Creates Its Own WorkScout runs every 4 hours and finds work for my fleet — stale tasks, downed services, unprocessed memory tags, alerts. Auto-creates Paperclip tickets and routes to the right agent. Here's the code.
-
Field Report How I Track 36 Agents' Cost Without Burning Agent CostA field report on the local-first cost loop I use to track 36 OpenClaw agents, catch spend spikes, and keep the watcher off the bill.
###april 202631 posts
-
Build Diary How I Open-Sourced 21 Scripts That Automate My AI Agent StackI run 36 agents 24/7 on a Mac Mini. Rather than babysitting them, I wrote scripts that let the fleet manage itself — self-healing, self-learning, zero-cost ops. Here's what they do.
-
Tools 160 AI Agent Templates: What I Learned From Deploying 36 AgentsBuilding agents from blank prompts is slow. After 36 production deployments, I catalogued what works into 160 tested templates — organized by use case, with the highest-value entries called out.
-
Operator Manifesto I Run 38 AI Agents — Here's What I Learned36 agents, 12 workspaces, one Mac Mini. What 18 months of building a production agent fleet actually taught me — what works, what fails, and what I'd do differently.
-
Methodology Multi-Agent AI Systems: What You Need to Know to Get StartedI run 36 agents. Getting there required understanding the core patterns first. Here's the mental model that makes multi-agent systems click — what they are, when to use them, and how to start.
-
Founder Will AI Replace My Job? A Task-Level Analysis of 42 RolesI spent 18 months mapping 42 roles against current AI capability. The answer isn't 'yes' or 'no' — it's task-level. Here's the data.
-
Methodology 7 Prompt Techniques That Actually Work (After Thousands of Hours)Most people treat AI prompts like search queries. After thousands of hours building production agents, here are the 7 techniques that consistently move the needle.
-
Methodology The 7 Golden Rules of Claude CodeNon-coder operating 36 agents built these rules from 18 months of production failures. Not theory. The exact patterns that separate sessions that ship from sessions that spiral.
-
Build Diary OpenClaw Evolution: 6 Phases of Building a Self-Healing Agent FleetBehind-the-scenes log of turning 36 agents from 'held together with duct tape' into self-healing infrastructure. Every phase paid for by a real failure.
-
Operator Manifesto The 10 Commandments of AI Infrastructure10 rules I follow operating 36 AI agents 24/7 across 5 companies. Every commandment paid for in production failure.
-
Architecture Multi-Agent Orchestration: Routing Work Across 36 AI AgentsHow work actually flows through a 36-agent fleet: 5 routing flows, channel rules, escalation paths. Not theory — production config.
-
Methodology AI Agent Autonomy Tiers: When to Auto-Approve, When to Wait4 tiers. 36 agents. A document that keeps them from breaking things they can't fix. What agents can do alone, what needs a human, and what's off-limits entirely.
-
Methodology Quick / Deep / SaaS / Overnight: How I Pick a Build Type Before Touching CodeThe decision that happens before any code runs. 4 build types, 4 checklists, 4 model choices. The wrong pick costs days. The right pick costs minutes.
-
Build Diary From One Agent to a Fleet: Building a Multi-Agent AI SystemStarted with one Claude session. Now I architect a production agent fleet across self-hosted infra and cloud platforms. Here's the architecture.
-
Methodology Killing AI Info Paralysis with a 3-Tier Reasoning System500+ skills. 50+ plugins. Where do you click? The system that reads your project and returns one starred next step — no round-robin, no hedging.
-
Tools 33 Claude Code Plugins You Should Know (and 5 You Shouldn't)A no-fluff breakdown of every plugin that ships with CC Commander — what each one does, when NOT to use it, and the 5 anti-patterns that burn tokens for no output.
-
Workflow Day Zero: Starting a Claude Code Project the Right WayFrom `npx commander` to first commit — the exact commands, CLAUDE.md template, and plan-mode session that prevent the 80% failure rate on Claude Code projects.
-
Workflow The Daily Claude Code Loop That Actually ShipsThe structured daily loop I run across 36 agents and 5 product brands — morning startup, session discipline, context rules, and the HANDOFF.md habit that prevents starting from zero every day.
-
Methodology The Iron Law of Debugging: No Fix Without Confirmed Root CauseEvery premature fix is a guess. Here's the investigation-first protocol — with the exact prompts, hypothesis-test loop, and operationalization steps — that eliminates repeat bugs.
-
Field Report Letting Claude Code Run Unsupervised for 8 Hours: A Field GuideWhat actually happens when you leave Claude Code running overnight — the setup, the checkpoints, the failure modes I've hit, and the security constraints that keep it from going sideways.
-
Founder The Non-Technical Founder's Guide to AI in 202650+ AI tools, every vendor promising transformation, none of them explaining what you're actually buying. Here's how to cut through it — no technical background required.
-
Provider Economics Frontier Model Convergence: When #1 vs #2 Doesn't Matter AnymoreThe top frontier models are now within 3-4 percentage points of each other on SWE-bench. When the gap is that small, model selection becomes a routing problem — not a vendor loyalty problem.
-
Founder The Hidden Costs of AI: What SMB Founders Don't SeeA restaurant owner was paying $847/month for AI tools. The real all-in cost was $1,400. Here's what the subscription stack hides — and the math that actually works.
-
Field Report I Let AI Manage My Smart Home for 2 Weeks1,119 entities. 30% offline at start. AI fixed them while I slept. The unglamorous truth about agents in the physical world.
-
Founder Where AI Actually Pays for Itself: 3 SMB Case StudiesNot every AI deployment delivers ROI. These three categories consistently do: customer support automation, ad copy generation, and data extraction. Real numbers from realistic operator scenarios.
-
Build Diary How I'm Monetizing an Open-Source Tool (Without Killing the OSS Spirit)Commander has 60 plugin skills, 88+ commands, growing install base. Every competing config kit is free. The 5 revenue models available — and which one I'm building first.
-
Build Diary WiFi Monetization: Turning Foot Traffic Into Loyal CustomersEvery day you offer free WiFi without a captive portal, a known customer walks out the door anonymous. Framework for turning foot traffic into a retargetable database — real numbers included.
-
Field Report Why WhatsApp Is the Future of Retail AuthenticationEmail forms are leaking. Facebook OAuth deprecated overnight. WhatsApp login is filling the gap — and the open rate numbers make the case obvious.
-
Tools 45 Claude Code Tips I Reference Every Week45 numbered tips from the BIBLE — the quick reference I built for Claude Code patterns that actually compound. Context management, git, multitasking, and the meta-skills most people skip.
-
Architecture The Architecture Behind My 36-Agent AI Fleet (Across 5 Companies)Not a concept piece. A working system: 36 agents, 5 brands, 24/7, at $2.48/agent/day. Gateway, orchestrator, specialist layer, free fleet, memory — all of it.
-
Architecture How I Organize Multi-Agent Output Without Losing My Mind36 agents produce a lot of files. Without a routing system, everything ends up in the wrong place and nothing is findable. Here's the file strategy that keeps it sane.
-
Provider Economics Running 200+ Free Models in Production (And When to Pay)Groq, Cerebras, Cloudflare Workers AI, OpenRouter. The escalation ladder I actually use, with cost data.