triggerdotdev / skills

trigger-setup bootstraps a project from nothing: installing the SDK, writing trigger.config.ts, scaffolding a first task in a /trigger directory, and setting TRIGGER_SECRET_KEY. trigger-tasks picks up from there with the actual task() and schemaTask() patterns — retries, queues and concurrency, debouncing and idempotency keys, run metadata, and cron scheduling. trigger-config goes deeper into the config file itself: build extensions for Prisma, Playwright, FFmpeg, and Python, plus environment-variable sync and telemetry. trigger-agents sits above tasks rather than beside them — prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer loops, and human-in-the-loop waitpoints — for LLM-driven pipelines rather than one specific API.

The other two are for the frontend and for the bill. trigger-realtime is the client-side half of the same system: React hooks such as useRealtimeRun and useRealtimeStream for showing a run's live progress or streamed output, plus minting scoped tokens so a browser can trigger a task directly. trigger-cost-savings is the odd one out — it needs Trigger.dev's own MCP tools already installed (npx trigger.dev@latest install-mcp) and uses them to read live run data for right-sizing machines, catching missing debounce or idempotency, and spotting polling loops that should be waitpoints instead.