Kavro

Think before you build.

AI coding agents are fast. Dangerously fast.

You give them a task. They jump straight to code. No research. No architecture. No tradeoffs. No long-term thinking. They produce something that compiles today and collapses in three weeks, and you're the one left debugging it at midnight.

This isn't a tool problem. It's a process problem.

Kavro is the fix.

Kavro is a universal AI engineering orchestration framework. It enforces a 7-phase Staff-level engineering workflow on top of any AI coding agent: Claude, Codex, Cursor, Windsurf, or anything else.

It is not a skill library. It is not a persona switcher. It is not a collection of prompts.

It is a process enforcement layer. A governing framework that wraps around the AI tool you already use and ensures it thinks like a senior engineer before it writes a single line of code.

Every Kavro session enforces this sequence. No shortcuts. No skipping.

Hard rule: No code is written before Phase 2 is complete. No phase is skipped. Ever.

The best engineers don't jump to solutions. They ask why before how. They document decisions. They think about the developer who inherits their code in 18 months. They design before they build.

AI agents don't do any of this by default. Kavro makes them.

→ Read the full philosophy

There are 232+ skills in the claude-skills repo. They teach Claude what to do. Kavro teaches Claude how to do anything.

Think of it this way:

• Skills are like tools in a toolbox
• Kavro is the process that decides when and how to use those tools

Before you ship any code, Kavro forces:

1. Research - understand the problem deeply
2. Architecture - design before building
3. Decomposition - break work into atomic tasks
4. Documentation - document as you go
5. Prompts - generate precise instructions per task
6. Selection - choose the right agent for each task
7. Governance - validate continuously

This is what senior engineers do naturally. Kavro makes AI agents do it too.

Kavro is built on the agentskills.io open standard - the same format adopted by both Anthropic and OpenAI. One framework, multiple adapters.

Auto-detects which tools you have installed and deploys the right adapter:

git clone https://github.com/a7medalyapany/kavro.git
cd kavro
bash scripts/install.sh

# Recommended: use the universal installer which bundles phase files
bash scripts/install.sh --claude

# Manual install (legacy): copy the adapter AND include the core/ tree
# so runtime references like core/phases/* resolve correctly after install
git clone https://github.com/a7medalyapany/kavro.git
cp -r kavro/adapters/claude ~/.claude/skills/kavro
# copy the core framework and phase files into the installed skill
cp -r kavro/core ~/.claude/skills/kavro/core

cd kavro
bash scripts/build.sh --claude
# Upload dist/kavro-claude.zip → Settings → Capabilities → Skills

cp -r kavro/adapters/codex ~/.agents/skills/kavro
# copy the core framework and phase files into the installed skill
cp -r kavro/core /.agents/skills/kavro/core

cp kavro/adapters/cursor/.cursorrules /path/to/your/project/.cursorrules

cp kavro/adapters/windsurf/rules.md /path/to/your/project/.windsurfrules
# Or paste into: Settings → AI Rules → Global Rules

See INSTALLATION.md for detailed per-tool instructions and troubleshooting.

The build and installer preserve the source layout under the installed

same in dev, in dist packages, and after installation. The canonical layout produced by bash scripts/build.sh --claude (dist/kavro-claude.zip) is:

kavro/
├─ SKILL.md
├─ agents/
│ └─ openai.yaml
├─ core/
│ ├─ KAVRO.md
│ └─ phases/
│ ├─ 01-research.md
│ └─ ...

Use the included validation scripts during development and in CI:

make test-paths       # scans the repo for broken or mixed path references
make validate-dist    # builds dist/ packages and validates contents
make test-install     # simulates install/uninstall in a temp HOME

Once installed, Kavro activates automatically when you start any engineering task. You don't invoke it explicitly - it triggers on intent.

Try asking your agent:

Design a multi-tenant SaaS backend with PostgreSQL and Redis.

Without Kavro: you get code.

With Kavro: you get a Research Summary → a Technical Blueprint with full decision documentation → a decomposed task list → implementation - in that exact order.

Kavro is an open standard framework. If you have improvements to a phase, a new adapter, or a better prompt pattern - open a PR.

See CONTRIBUTING.md for guidelines.

See CHANGELOG.md for version history.

MIT LICENSE - use it, fork it, build on it.

Kavro - Think before you build.