The skills my team runs
in production.
10 Claude Code skills: onboard any repo to agent-driven dev, give every agent its own stack, make the work verifiable, build loops that keep going.
/plugin marketplace add AI-Builder-Club/skills /plugin install skills@ai-builder-club
One plugin, all 10 skills. Works in any repo.
30+ loops
running our 3 brands on these skills
#1 clicked page
on our site 7 days after one loop shipped it (seo-growth)
-4 to 95 karma
on Reddit in 7 days, run end to end by one loop
A loop is an agent that owns a workstream
It wakes on a trigger, does the work, and writes what it found into a shared, file-based memory. Next run it reads that memory and keeps going. Building one takes four ingredients. This plugin ships two of them.
01
Triggers
A cron, a webhook, an incident, or another agent wakes the loop at the right time.
You wire02
Shared memory
The file and logging structure every loop reads and writes. One brain, many loops.
Ships here03
Tools & connectors
Your skills and MCPs, so the agent can do real work in your systems.
You bring04
Codebase harness
So the agent can run, test, and verify its own work before it ships.
Ships hereTwo entry points
Everything else in the plugin hangs off these two commands.
Make a repo agent-ready
Run it in the code repo your agents work in, so they can run, test, and verify their own work. It orchestrates the four harness skills below and pulls in only what the repo needs.
Read the skill/new-loopBuild your shared brain
Run it where your agent's memory should live. First run bootstraps the knowledge base, then scaffolds the loop and does one real test run. Run it again any time to add another loop.
Read the skill10 skills, grouped by job
Every skill is a folder you can read before you run it. No black boxes.
Codebase harness
make a repo agent-readyOnboarding a repo to agent-driven dev. The master skill that orchestrates the four below.
↗dev-local-setupA one-command local dev stack. scripts/dev-local.sh up, and the agent has an app to test against.
↗e2e-setupThe repo has no real e2e? Add a per-PR test gate agents must pass before shipping.
↗crabbox-setupLoops shipping code in parallel each get an isolated cloud stack. One laptop can't run N.
↗verifier-setupScaffolds a repo-specific /verify skill: a fresh sub-agent drives the app and attaches screenshot proof to the PR.
Loops
the shared knowledge baseUtilities
standalone, use anywhereAudit your CLAUDE.md, docs, and skills against Anthropic's context-engineering guidance. Finds conflicts, stale facts, and overconstraint.
↗open-agent-teamsDelegate work to any CLI agent (claude, codex, aider) in a detached tmux session with a race-safe done signal.
↗seo-growthWhere to point SEO effort: a cold-start playbook and a with-data playbook, distilled from two sites we run in production.
↗visual-flow-gifTurn an article, workflow, or architecture into a static PNG plus an animated GIF flow diagram.
New skills ship to the newsletter first
When we add a skill or change how we run our loops, subscribers hear about it before anyone else.
Free AI Builder Newsletter
Weekly guides on AI tools & builder strategies.
Watch
The loop engineer: how to design compounding agent loops
The full walkthrough of the concept and how my team designs loops that read and write one shared brain.
This is the setup my team runs. Take it.
Open source, readable, and built to be pulled apart. Star it, install it, then come tell us what your loops shipped.