AI Engineering

Architecture designs greenfield systems by mapping functionality flows, components, APIs, dependencies, and observability. Use it for new systems or major multi-component features; for existing code with shallow modules, use codebase-hygiene instead.

Cc10x Router routes code requests into build, debug, review, plan, triage, or orient workflows. Use it when a request needs immediate orchestration, task execution, validation, and persistence.

Codebase Design defines vocabulary for shaping modules with small interfaces and deep implementations. Use it when designing seams, improving testability, or deciding whether an abstraction is shallow or deep.

Domain Modeling actively shapes a project's glossary and domain terms by challenging fuzzy language, testing edge cases, and updating CONTEXT.md as the model evolves. Use it when defining or refining concepts, and in read-only mode to apply the project's established vocabulary.

ElevenLabs converts text to speech, generates sound effects and dialogue, transcribes audio or video, and manages voices and quota. Use it for narration, voiceovers, transcription, voice changing, or synthetic voice design.

Exploration turns rough ideas into validated designs through collaborative interview, or creates throwaway spike code to answer one design question. It is used when an agent needs to clarify scope before planning or prototype a risky choice.

HeyGen generates AI avatar videos, translates and lip-syncs videos, synthesizes speech, and browses or clones voices and avatars. Use it when creating or checking HeyGen video and voice jobs.

Nexus Mapper generates a persistent .nexus-map knowledge base that summarizes a codebase’s architecture, dependencies, concepts, and git hotspots. Use it when onboarding to an unfamiliar repo or preparing a large refactor with AI assistance.

Nexus Query answers code-structure questions from AST data, such as who imports a module, what depends on an interface, and how far a change will ripple. Use it before refactors, interface changes, or when estimating risk in large legacy codebases.

Perplexity asks web-grounded questions with citations, runs and polls deep research jobs, lists available models, and searches the web. Use it for current answers or ranked search results.

Plan Review Gate checks saved plans and decision RFCs for feasibility, completeness, and alignment before execution. It is used after non-trivial planning when a fail-closed review should block unsafe or incomplete work.

Planning distills tasks into durable execution plans and decision RFCs. Use it to break work into bite-sized tasks, define validation, manage risk, and map functionality flow.