Codexpiator Router

Codexpiator is a set of small, single-purpose skills covering the full span of building and shipping software: frontend, backend, security, testing, architecture, devops, git/DX, and AI integration. This router's only job is to classify an incoming request and point at the right skill (and, inside it, the right file) — never to answer from here directly with duplicated content.

Classification table

Domain	Dispatch to	Example triggers
Frontend (UI, components, state, styling, forms, routing, a11y)	codexpiator-frontend	"how should I structure this component", "state management approach", "form validation", "is this accessible"
Backend (APIs, data, auth, jobs, caching, resilience)	codexpiator-backend	"design this endpoint", "should I cache this", "how to model this data", "background job for emails"
Security (secure coding, authn/authz, injection, secrets, deps)	codexpiator-security	"is this safe", "how do I store passwords", "sql injection risk", "secrets management"
Testing & QA (strategy, TDD, mocking, CI gates)	codexpiator-testing-qa	"what should I test", "TDD for this feature", "flaky test", "what blocks merge"
Architecture (structure, patterns, contracts, scaling)	codexpiator-architecture	"how to structure this project", "monolith or microservices", "API versioning", "will this scale"
DevOps (CI/CD, environments, containers, observability, incidents)	codexpiator-devops	"set up CI/CD", "dockerize this", "how to monitor this", "rollback plan"
Git / DX (workflow, commits, PRs, review, docs, onboarding)	codexpiator-dx-git	"branching strategy", "write a commit message", "review this PR", "onboarding docs"
AI integration (LLM patterns, prompting, agent safety, MCP)	codexpiator-ai-integration	"integrate an LLM", "prompt engineering", "is this agent safe", "which MCP server"

Invoke the matching skill (through the environment's actual skill- invocation mechanism, by name — e.g. codexpiator-frontend — the same way a superpowers:* skill gets invoked) rather than opening its SKILL.md with a file-read tool. Invoking it properly lets it activate and point itself at its own specific resource file(s); don't answer from this table's one-line descriptions alone, and don't copy a topic skill's content into this router.

A request can span more than one domain (e.g. "add a login feature" touches frontend, backend, and security). In that case, consult each relevant skill for its slice rather than picking just one.

Long or complex work

For anything substantial (multi-step, many files, spans multiple commits), see shared/long-task-memory-and-superpowers.md — write resumable progress memory to a gitignored project directory, and use superpowers:* skills (writing-plans, executing-plans, systematic-debugging, TDD, etc.) when available instead of improvising the process.

Ask when useful; suggest an audit after significant work

See shared/collaboration-and-audit-practice.md — ask a targeted clarifying question (via AskUserQuestion when available) when a request has real ambiguity, and suggest /codexpiator-audit after completing significant work. Every codexpiator-* skill follows this.

Before answering anything design- or security-audit-shaped

If the request involves visual design, motion/animation, a broad UI polish pass, redesigning an existing UI, or a deep security audit, read shared/external-skills-registry.md first. It has the exact procedure for checking whether a more specialized external skill is available, and the fallback to use when it isn't. Don't skip this check just because codexpiator-frontend or codexpiator-security also has relevant content — the registry decides which one leads.