AI Engineering

AI Product Development helps ship production-grade AI features with LLM integration, RAG, prompt engineering, streaming, safety guardrails, and cost control. It is used when building AI products that need structured outputs, validation, and user trust.

API Design Principles helps you design and review REST and GraphQL APIs with clear resources, methods, versioning, pagination, and error handling. Use it when defining API standards or improving existing endpoints.

Arb Review generates an 11-slide architecture review board deck from PRD, architecture maps, and project context. Use it to prepare a sign-off presentation with diagrams, risks, NFRs, and a decision ask.

Brainstorming helps turn an idea into a design and spec before implementation. Use it to clarify requirements, explore approaches, and get approval before any code is written.

CLI Anything wraps a command-line tool in a documented agent contract with JSON output and error handling. Use it when you want an AI agent to call a CLI reliably and parse its results.

Cobbler invokes Elves’ default orchestration model for fitted answers and agent routing. Use it when a user types /cobbler or needs Cobbler coordination in Claude Code.

Cobbler Mode turns on Elves Cobbler Mode for the current Claude Code thread. Use it to keep follow-up prompts mediated by Cobbler after /cobbler-mode or when the user wants Cobbler by default.

CoCo turns Claude Code into a PM-style control layer with dashboard activation, project context, todos, search, and health checks. Use it when you want one interface for product knowledge, decisions, and session status.

Coco Loop turns a plain-language goal into a governed, propose-only autonomous loop. Use it when you want a bounded session that proposes fixes, runs under a constitution, and never commits on its own.

Council is a compatibility alias for Elves Cobbler orchestration in Claude Code. Use it when a user types /council to route work through the default coordination flow.

Devin Remote Task starts and boundedly follows a remote Devin development task from Claude Code. Use it when a user runs /devin with implementation instructions and you need to launch the task, keep the repo as the working directory, and report follow links.

Dispatching Parallel Agents is used when 2+ independent problems can be investigated without shared state or sequential dependencies. It helps split work into focused agents that run concurrently and report back for integration.