inngest/inngest-skills
6 skills14k installs138 starsGitHub
Install
npx skills add https://github.com/inngest/inngest-skillsSkills in this repo
1Inngest EventsInngest Events is a workflow-oriented agent skill for designing the event layer that powers durable background jobs on Inngest. Solo builders decoupling microservices, billing webhooks, or internal notifications learn the required name-and-data contract, optional deduplication identifiers with a twenty-four-hour window, timestamp semantics, and versioning hooks so functions remain safe under at-least-once delivery. The skill walks fan-out architectures where a single business occurrence—like order.placed—fans to fulfillment, analytics, and email handlers without tight coupling, and it surfaces system events such as inngest/function.failed for compensating flows. Although examples skew TypeScript, the conceptual guidance transfers to Python and Go stacks via official Inngest documentation links, which matters when your agent spans a Node API today and workers tomorrow. Reach for this skill during Build when wiring Stripe or custom producers, then again in Ship and Operate when hardening retries and failure telemetry. Correct event design upfront prevents duplicate charges, double emails, and orphaned partial workflows that are painful to debug in production.2.7kinstalls2Inngest StepsInngest-steps is the procedural reference for building durable workflows with Inngest step APIs. Solo builders use it when a process must sleep for hours, wait for a human approval event, poll an external API without losing progress on deploy or crash, fan out parallel step.run work, or memoize expensive calls across retries. The skill’s central discipline is that every step invocation re-executes the function from the top, so API calls, randomness, and database reads must live inside steps—not in bare function scope—or retries will duplicate side effects. It enumerates step.sleep and wait helpers, cross-function invoke, AI helpers, and loop/parallel patterns while noting hard limits of 1,000 steps and 4MB serialized step payload per function. Intermediate to advanced because misuse causes subtle double-charging or stuck workflows. Canonical shelf is build/backend; the same patterns matter in operate when extending live jobs.2.7kinstalls3Inngest Durable FunctionsInngest-durable-functions teaches solo builders how to use Inngest checkpointing to shrink latency in durable, event-driven workflows. Without checkpointing, each step.run typically implies an orchestration round-trip on the order of tens of milliseconds, which stacks badly in chatty pipelines. Checkpointing keeps steps executing eagerly on your runtime while Inngest persists checkpoints for crash safety and replay. The skill walks through side-by-side TypeScript patterns so an agent does not blindly copy the slow model into latency-sensitive paths. It fits SaaS and API backends that already emit Inngest events and need production-grade durability without sacrificing responsiveness. Intermediate complexity: you must understand Inngest functions, triggers, and when checkpoint semantics trade off against strict step isolation. Use during backend implementation and when tuning operate-phase jobs that users feel in real time.2.7kinstalls4Inngest Setupinngest-setup teaches solo builders how to bolt Inngest’s durable execution layer onto a TypeScript codebase from zero. The skill walks package installation, creation of a shared Inngest client with a stable application id, and the framework-specific serve or connect-as-worker wiring that lets functions register with Inngest Cloud or the local dev server. It targets the common indie SaaS pain points: webhook handlers that must retry safely, background jobs that outlive a single HTTP request, cron-like schedules, and longer workflows that should not die when a server restarts. Prerequisites are explicit—modern Node, an existing TypeScript project, and a package manager—so agents do not hallucinate config for the wrong runtime. The document flags that Inngest v4 defaults to Cloud mode and that local development requires deliberate configuration, which prevents silent “works on my laptop” failures. While Python and Go are out of scope for the step-by-step commands, the skill still orients polyglot teams to official docs. For Prism users in Cursor or Claude Code, this is the first skill to invoke before writing Inngest functions, steps, or event schemas.2.7kinstalls5Inngest Middlewareinngest-middleware is a reference skill for solo builders running durable functions on Inngest who need correct imports and options for client-side middleware. It clarifies a common pitfall: only dependencyInjectionMiddleware ships from the core inngest package, while encryption and Sentry integrations require dedicated npm packages with their own install steps. The encryption section spells out what is encrypted by default—step payloads, function output, and the configurable event field—and documents operational options such as eventEncryptionField, decrypt-only mode for migrations, and fallback keys for rotation. Sentry wiring is pointed to the middleware-sentry package with the same separation of concerns. Use this when an agent is scaffolding or reviewing your Inngest client bootstrap so you do not import nonexistent symbols or leak unencrypted event bodies in regulated or multi-tenant SaaS backends.2.6kinstalls6Inngest Flow ControlInngest Flow Control is an agent skill that teaches solo builders how to tame serverless job traffic with Inngest’s concurrency, throttling, rate limits, debounce, singleton, batching, and priority options. It is written for TypeScript backends but notes that core concepts transfer to Python and Go via Inngest documentation. Reach for it when queues spike, duplicate webhooks fire, or expensive steps need caps without freezing entire function runs. The skill’s decision guide turns vague reliability worries into concrete configuration choices—for example knowing debounce waits for quiet periods while rate limiting hard-stops bursts. That clarity helps one-person teams ship durable async workflows without hiring a platform engineer. Pair it with your function definitions during build and revisit limits after launch when real traffic patterns appear.614installs