auk

npx auk-develop init && npx auk-develop generate

The problem

Every AI coding tool needs context about your codebase — and today you write it by hand:

• It's generic. "Use TypeScript. Write clean code." Your agent already knew that.
• It rots. You migrated from axios to fetch six months ago. Your CLAUDE.md still says axios. Your AI confidently writes axios.
• It's fragmented. CLAUDE.md for Claude Code, .cursor/rules for Cursor, .github/copilot-instructions.md for Copilot, .windsurfrules, .aider.conf.yml, GEMINI.md… all drifting apart.
• It has no memory. Your codebase made hundreds of architectural decisions. Nobody wrote down why. Your AI can't tell an intentional pattern from an accident.

Existing tools visualize your code (knowledge graphs, dashboards) or manage hand-written rules. Nobody builds the context from the code. auk does.

What auk generates

Not this:

"Use TypeScript. Follow existing patterns. Write clean code."

This:

🔴 Error handling uses the Result<T,E> pattern from src/types/result.ts — never throw in the service layer, always return Result. (94% of services follow this — verified)

🔴 Services never import from Controllers. 3 existing violations are known tech debt — don't add more. (verified on every commit)

🟡 UserService has 11 connections and is becoming a god object — don't add more responsibilities.

Every rule carries evidence (file:line references), a confidence score, and a machine-verifiable claim — so it can be proven true or flagged as rot, automatically.

Quick start

cd your-project
npx auk-develop init        # detect your stack, create .auk/
npx auk-develop generate    # analyze code → rules → all agent formats

That's it. Now every AI tool you use reads context derived from your real code.

Claude Code plugin (one-command deep pass)

The CLI extracts the facts (parsing, call graph, clusters, conventions). The depth — rewriting each rule into sharp, codebase-specific guidance — is best done by an agent that can read your code. The Claude Code plugin makes that one command: your agent does the semantic pass, no API keys, no copy-paste.

/plugin marketplace add sandeepshekhar26/develop-anything
/plugin install auk@auk

Then, in any project:

/auk:run        # full pipeline: analyze → agent deep-enhances every rule → compile every agent file
/auk:enhance    # just the deep-enhance pass on existing rules, then recompile

/auk:run runs the analysis, reads .auk/prompts/* plus the cited code, rewrites the rules, applies them, and regenerates CLAUDE.md / AGENTS.md / .cursor/rules / copilot-instructions — all in one keystroke. If a command doesn't appear, run /reload-plugins.

Other agents (Cursor, Copilot, Codex, Gemini CLI…) can run the same agent-agnostic loop — see plugins/README.md.

The commands