Free, dated, no fluff. The same playbooks every ClaudeFarm crop is built on — published as we ship.
One hub is one PC's disk — so we pooled several. A consistent-hash ring over self + peers, sha256(docId) to pick the owning node, secret-authed local-only endpoints, and stripe (RAID-0 capacity) vs mirror (RAID-1 redundancy). A real NAS alternative over your own Wi-Fi, still a zero-dependency Node script.
Six Claude agents in parallel, one shared interface contract, zero integration conflicts. Editor, real OOXML DOCX export, PDF import, BYOK Claude skills, a zero-dependency LAN document hub, and Stripe licensing — all live same-day at writehive.netlify.app.
$5/M input, $0.50/M cached, $25/M output — verified June 2026. Worked cost examples, why prompt caching cuts a workflow bill ~90%, and the true cost of a 200k-token codebase day (we watched it drop from $208 to $37).
A skill is a SKILL.md with frontmatter that lives in ~/.claude/skills/. What a skill is, where they live, how triggering works, how to write and test your own, common mistakes, and the 5 worth installing first.
The Skills ecosystem just hit 15,134 repos. Most are dead weight on Opus 4.7. Here are the 12 we run in production every day — ranked by what actually ships value, with honest "skip this if…" notes on each.
Eight patterns that justify the token spend, three that backfire. When 1M context replaces RAG, when it doesn't, and how to budget tokens without burning $40 per query.
New tokenizer, file-system memory tool, 2576px vision, Fast Mode. Step-by-step migration with code samples — plus the three gotchas that broke our agents in week one.
Anthropic's public-beta "dreaming" lets agents review past sessions overnight and self-improve. Here's what it does well, where it fails, and how to wire it into a multi-agent workflow.