codebound

# codebound codebound is a router skill for Vercel-native code-first GTM engineering on Vercel + Supabase + Vercel Workflows + Eve + AI SDK + Next.js. ## Route - For Codebound product doctrine and vocabulary, read `references/philosophy.md` and `references/glossary.md` only when the branch needs shared definitions. - For paid resources, destructive operations, production writes, CRM mutations, AI-generated claims, browser marketplace flows, or local-vs-remote environment risk, read `references/safety.md` and follow its gates. - If the user asks about readiness, doctor, prerequisites, setup checks, auth checks, CLI versions, scaffold, initialize, resume, repair, destroy, deploy, upgrade, migrate, scaffold version, serve locally, spin up localhost, or local development environment, read `references/project.md` and follow it. - If the user asks about schemas, Supabase tables, internal schemas, Studio views, view CRUD, existing data, dedupe, identity, normalized facts, provenance, freshness, or private database access, read `references/data.md` and follow it. - If the user asks about Vercel-native stack choices, Workflows, Eve, AI SDK, AI Gateway, Connect, Sandbox, Queues, or migration off Inngest, read `references/vercel-stack.md` and follow it. - If the user asks to create or change a GTM workflow, source, CSV import, webhook intake, scheduled source, Vercel Workflow, local test, run, replay, preview trigger, or production trigger, read `references/workflows.md` and follow it. - If the user asks about Eve, agents, agent tools, agent skills, subagents, channels, schedules, conversational GTM assistants, Slack agents, or human-in-the-loop operators, read `references/agents.md` and follow it. - If the user asks about AI SDK, AI Gateway, model calls, structured outputs, streaming, AI research, model routing, or model costs, read `references/ai.md` and follow it. - If the user asks about Vercel Connect, OAuth, scoped auth, Slack, GitHub, Salesforce, Notion, Linear, or connected external systems, read `references/connect.md` and follow it. - If the user asks about sandboxed execution, generated scripts, untrusted code, scraping experiments, or code-running agents, read `references/sandbox.md` and follow it. - If the user asks about queues, fanout, consumer groups, message routing, visibility timeout, dedupe, or lower-level workflow routing, read `references/queues.md` and follow it. - If the user asks about enrichment providers, external APIs, provider adapters, AI research, waterfalls, credits, provider costs, provider benchmarks, or normalized provider outputs, read `references/providers.md` and follow it. - If the user asks about only-run-if behavior, auto-run behavior, schedules, delays, rate limits, concurrency, budgets, dry runs, kill switches, retries, replay safety, or backfills, read `references/rules.md` and follow it. - If the user asks about CRM sync, HubSpot, Salesforce, sequencers, Slack, exports, outbound webhooks, warehouse writes, ads audiences, human review queues, or side-effecting operations, read `references/actions.md` and follow it. - If multiple intents appear, run readiness/checking before scaffolding unless the user clearly asks otherwise. - If the requested route is not implemented yet, say that codebound does not implement that route yet. Do not direct users to legacy separate slash commands. The public entrypoint is `/codebound` with natural-language intent.