auk works standalone (npx -y auk-develop generate), but the deep pass —
rewriting each mined rule into sharp, codebase-specific guidance — is best done
by an agent that can read your code. These integrations make that one command.
The split is deliberate:
- Structural facts (parsing, call graph, clusters, conventions) come from the
deterministic
auk-developCLI — fast, free, repeatable. - Semantic depth (why a rule exists, what breaks if violated, the real file and type names) comes from your agent during the deep pass — no API keys, no copy-paste.
Install from this repo as a plugin marketplace:
/plugin marketplace add sandeepshekhar26/develop-anything
/plugin install auk
Then, in any project:
/auk:run # full pipeline: analyze → deep-enhance every rule → compile all files
/auk:enhance # just the deep-enhance pass on existing rules, then compile
/auk:run runs auk generate --emit-prompts, the agent reads .auk/prompts/* and
the cited code, writes the enhanced rules back, applies them, and compiles
CLAUDE.md / AGENTS.md / .cursor/rules / copilot-instructions — all in one go.
You can also register the MCP server for live queries:
{ "mcpServers": { "auk": { "command": "npx", "args": ["-y", "auk-develop", "mcp"] } } }and ask the agent to use get_enhancement_tasks + apply_enhancements.
The deep pass is agent-agnostic. Add this to your agent's rules file
(.cursor/rules/auk.mdc, .github/copilot-instructions.md, GEMINI.md, …):
When asked to set up or refresh auk context: run
npx -y auk-develop generate --no-compile --emit-prompts, read every.auk/prompts/enhance-rules-NN.md, open the cited code, and rewrite each rule to be specific to this codebase (real dirs, types, frameworks) with a rationale. Write eachenhance-rules-NN.response.jsonper the embedded schema, runnpx -y auk-develop enhance --apply <file>for each, thennpx -y auk-develop compile.
Prompts in, schema-validated JSON out — the same loop the Claude Code command automates.