Free guides to get more out of Claude Code, and a ready-made harness of 2,104 skills and 7 self-improving subsystems for when you'd rather not build it all yourself.
security-review/SKILL.md---name: security-review
description: Use when adding auth,
user input, secrets or new endpoints.
---## Checklist
- Validate input at the boundary
- Rate-limit auth routes
- No secrets in source
- Errors don't leak internals
>add rate limiting to /login● skill security-review
● memory: use lib/cache.ts
● hook pre-edit: tests first
src/routes/login.ts +18−2
tests/login.rate.ts +41✓ 12 passed· review clean
2,104skills
196commands
80agents
7subsystems
How it works
Install it. Open Claude Code. It's already better.
Keep scrolling: one install, then Claude Code runs a skill, recalls last week and hands off to agents.
InstallOne command. Your old setup is backed up first.
LaunchClaude Code opens with the harness on board.
Run a skill/prompt-engineer loads only when you call it.
Memory + agentsIt recalls your rules and splits the work.
~/shop — zshexample session
What you get
Everything your Claude Code is missing.
Not a plugin. A complete setup for Claude Code: skills, memory and self-healing in every session, your tools connected safely, a check that it really works on your machine, and an evolution cycle you run when you choose.
loaded by context
typescript-procore
systematic-debuggingcore
react-expertlibrary
security-reviewcore
postgres-prolibrary
test-driven-developmentcore
kubernetes-specialistlibrary
golang-prolibrary
frontend-designcore
deep-researchcore
api-designcore
rust-engineerlibrary
gsap-scrolltriggercore
motion-designcore
design-dnacore
typescript-procore
systematic-debuggingcore
react-expertlibrary
security-reviewcore
postgres-prolibrary
test-driven-developmentcore
kubernetes-specialistlibrary
golang-prolibrary
frontend-designcore
deep-researchcore
api-designcore
rust-engineerlibrary
gsap-scrolltriggercore
motion-designcore
design-dnacore
01 skills2,104
198 core. 1,906 on demand.
Core skills switch on by context. The library is searchable when you need it: TypeScript to Rust, React to Kubernetes, GSAP animation to design systems.
02 commands196 · 80 agents
One command away.
36 namespaces covering TDD, review, deploy and research, with 80 specialist agents behind them.
memory · ~/shop● on
03 memoryon
Claude remembers last week.
claude-mem records every tool use, compresses it and brings it back next session.
evolution storegen 41
04 evolutionby selection
Winning approaches get stronger.
Every session is logged to the evolution store. When you run an evolve cycle, strategy weights shift by selection, so what works for you is used more and what doesn't fades.
✓ all 412 checks passing
05 self-healingsentinel
Broken? Quarantined. Repaired.
Each session starts with a quick check of hooks, store files and MCP servers. /harness:sentinel checks every skill and config file, then isolates and repairs what breaks.
connect services● stays on this computer
GitHubsigned in through your browser
Exakey checked, in the key store
Geminikey checked, in the key store
Stripesign in later
06 serviceskeys stay local
Connects your tools. Keeps the keys.
The installer signs you in to GitHub and asks for the keys your skills use, each with why it's needed. Keys go in your computer's own key store, never in a settings file and never to us.
$ overclaud doctor --device
✓ settings ~/.claude
✓ hooks 29 runs, 6 events
✓ safety blocked in-session
✓ memory on
✓ works on this computer
07 device checkafter every update
Proves it works on your machine.
After installing and after each daily update, a short real Claude Code session checks the hooks fire and the safety hook blocks. If something's off, Claude applies a tested fix. A failed update rolls itself back.
release 1.4.1● signed
files7,621 checked before install
pluginspinned to reviewed versions
GitHubGitHub's own MCP server
tokensnone in config files
08 securitysigned · pinned
Signed, checked and pinned.
Every release is signed and every file is checked before it installs. Plugins are pinned to versions we reviewed, GitHub runs through GitHub's own server, and no tokens sit in config files.
The engine
7 subsystems that review and tune your setup.
Each one feeds the next. Hooks record every session, and the loop runs when you start it with /harness:evolve.01 / 07
01ORACLEStrategizes
Reads the evolution store, weighs lessons by how recent they are, and recommends what to do next.
"tests first on payments"2d
"small diffs review faster"9d
"parallel agents on docs"40d
→ next: run /tdd before editing billing/
02SCOUTDiscovers
Searches GitHub, curated lists and tracked sources for new skills, MCP servers and plugins.
stripe-mcpmcp★ 2.1k
playwright-skillskill★ 870
pg-explainskill★ 412
context-trimplugin★ 260
03FORGEIntegrates
Checks what SCOUT finds for conflicts and risk, then installs approved items with rollback.
✓No name conflicts
✓Risk score: low
✓Rollback point saved
✓Installed stripe-mcp
04CRUCIBLEOptimizes
Runs experiments on your config. Winning strategies are promoted, losing ones dropped.
A · plan first0.72
B · code first0.48
A promoted · B dropped · 38 sessions
05MIRRORReflects
Reads session telemetry for usage patterns, dormant skills and recurring gaps.
14 dormant skillsgap: no k8s rules
06SENTINELHeals
Validates every skill, hook and config file. Quarantines what breaks and repairs it.
✓skills/api-design/SKILL.md
✓settings.json
✗✓hooks/pre-edit.shexit 127repaired
✓commands/deploy.md
07INTEGRATEAbsorbs
A four-phase pipeline that pulls whole repos and architectures into your setup.
Clonerepo
Analysepatterns
Adaptyour rules
Mergewith tests
Start here
Three ways to get more from Claude Code.
Learn it yourself for free, install the whole setup in one command, or set it up with me.
I'm Taner Hassan. Overclaud is the Claude Code setup I use daily for research, development and review, on my own products and on client work.
Before software I ran operations: institutional operations at a platform with $10B under management, an OTC desk I founded, and three years as COO of an angel-backed startup. Today I'm CTO and co-founder of Crumbless AI, and I build agentic systems for founders and operations teams.
Rolling out across a company? A live 2-hour workshop, licences for everyone and a playbook, from $1,499. Talk to us.
Field notes
Not ready to buy? Learn the craft first.
Practical notes on getting more out of Claude Code: skills, hooks, MCP and agent workflows, plus what changes in each Overclaud release. Free, by email.
FAQ
Questions.
What is Overclaud exactly?
A complete configuration for Claude Code, the CLI by Anthropic. It replaces your ~/.claude folder with 2,104 skills, 196 commands, 80 agents, lifecycle hooks and a self-improving harness. Your existing setup is backed up first.
Do I need a Claude subscription?
Yes. Claude Code needs a Claude Pro or Max subscription from Anthropic. Overclaud is the configuration layer. Anthropic provides the AI.
What does self-improving mean?
Seven subsystems form a feedback loop. Hooks record what happens in each session. When you run /harness:evolve, MIRROR analyses what you use, SCOUT finds new tools, FORGE integrates them and CRUCIBLE tunes your config. SENTINEL checks health at the start of each session and repairs what breaks.
What are lifetime updates?
Pay once and get every future version: new skills, new commands, and compatibility fixes when Claude Code changes. Overclaud updates itself once a day, and rolls back automatically if the install checks fail. To update now, run overclaud update.
Can I use one licence on several devices?
Each licence activates on one device. Add a licence per machine, or email support@overclaud.dev to move yours to a new one.
What if Claude Code changes and breaks it?
After every daily update a short real Claude Code session checks that the hooks fire and the safety hook blocks; if something is off, Claude applies a tested fix, and an update that fails its checks rolls itself back. For breaking changes in Claude Code we ship compatibility updates, which arrive with the daily update, or run overclaud update to get them now.
Which languages and frameworks are covered?
TypeScript, JavaScript, Python, Go, Rust, Java, C#, C++, Kotlin, Swift, PHP and SQL. React, Next.js, Vue, Angular, Django, FastAPI, NestJS, Rails, Laravel, Spring Boot, Flutter and more, plus Kubernetes, Terraform, Docker, AI/ML, security and design.
Where are my API keys stored?
Overclaud asks for the sign-ins and API keys it can use, and says why each is needed and where to get it. Keys go in your operating system's key store (Windows Credential Manager, macOS Keychain, Linux Secret Service). They are checked with a harmless request when you enter them, never written to settings files, and never sent to Overclaud. Manage them with overclaud keys.
Can I buy for my whole team?
Yes. Team packs give each person their own key at a lower per-seat price: 5 seats $245, 10 for $390, 25 for $725. All keys arrive by email to hand out.
Is there a refund policy?
Overclaud is delivered instantly, so purchases are generally non-refundable. If your key genuinely doesn't work and we can't fix it, email us within 14 days for a full refund. Your statutory rights under the UK Consumer Rights Act are unaffected. See the terms.
Stop configuring. Start evolving.
Your Claude Code is running at a fraction of what it can do. One install fixes that.
