The winner of an Anthropic hackathon open sourced his entire Claude Code setup and this is absolute f*cking gold. 68 subagents, 286 skills, 94 commands, MIT license. ECC turns one Claude Code assistant into a staffed engineering org. The plan lands before the build, the failing test lands before... the fix, and every change gets read again by a context that never watched it get written. • Who does what > Planning -> give it one sentence, get a plan you approve before any code exists. > Review -> reads your diff cold, with a separate reviewer per language. > Build repair -> one fixer per toolchain, PyTorch and CUDA included. > Security -> an OWASP sweep, plus a scanner looking for injection holes in your agent config. > Architecture -> kills design mistakes while they're still cheap, long before migrations. > Domain work -> database queries, ML pipelines, e2e tests, docs. That security pair is what almost nobody bothers to set up. An outside OWASP audit runs four figures and a week of waiting this one finishes on your branch by lunch. • What the skills cover > Testing -> tdd-workflow walks red to green, eval-harness sits on top. > Language packs -> Python, Go, Rust, C++, Django, Laravel, Spring Boot, Next.js. > Context -> search-first reads the docs before writing, iterative-retrieval keeps your repo from flooding the window. > Shipping -> Docker, CI/CD, health checks, rollbacks, migrations. > Beyond code -> writing in your voice, market research, pitch decks. Fork it, cut it down, run your own version tomorrow. Start with one plan and one rules pack switching on all 286 skills at once is the fastest route to a worse setup. Whoever wires this in over a weekend spends next quarter reviewing work instead of typing it.show more

68 subagents in one claude code setup is fucking wild!

search-first legit saves my tdd runs from junk context overflow

68 subagents isn't just volume; it's specialization. The real gold is how Claude Code routes context so those agents don’t step on each other’s toes while the main thread stays clean. Organized chaos at its finest. 💡

上下文隔离的 review 确实能减少盲点，但 68 个 subagent 全量调度的 token 成本在大项目里很难忽略

Start with one plan, one reviewer, and one repair loop, then measure escaped bugs and review time before adding more skills. More agents only help when each has a clear handoff and failure boundary.

one plan and one rules pack is the usable line

Saved this info

fresh context makes review actually independent

68 subagents feels like the kind of number that sounds better than it runs.

the cold reviewer on the diff is the one i want

That absolute level of detail feels like a cheat code for solo devs turning into a mini-team

Sixty-eight subagents is a staffed org on paper. Coordination is the real bottleneck: plans diverge, usage limits scale with thread count, and nobody owns the merge. A harness that plans first still needs one owner.

ter until one acceptance test kills a bad plan. Whats the first check that must fail before any subagent is allowed to write code?

ECC is a real repo. This post is a staffing fantasy. 68 subagents and 286 skills are markdown jobs, not a company. One model is still doing the work. Load the whole catalog and you spend ~27k tokens introducing the toolbox before a line of product code exists. The post even says turning it all on makes the setup worse, then sells the weekend install anyway. The OWASP line is the tell. A config scanner that looks for leaked keys and sloppy hooks is not a four-figure security audit that finishes by lunch. Different object, same scare price. Fork it. Keep a planner, a cold reviewer, and the rules for your stack. Leave the other 280 skills on the shelf. A harness is a folder you edit. It is not a department that showed up for free.

This is very useful information!

This Claude Code setup is amazing. Who will try it out first?

switching all 286 skills on at once sounds cooked ngl

found what ill read todays evening

At this point Claude Code doesn’t need another prompt, it needs an HR department.