Healthcheck

Service & dependency signals

!grep -rilE "express|fastify|koa|nestjs|flask|fastapi|django|gin|echo|actix|rails|spring" package.json pyproject.toml go.mod Gemfile 2>/dev/null | head

!grep -rilE "createConnection|DataSource|pg|mongoose|redis|prisma|sqlalchemy|database_url" --include='*.*' . 2>/dev/null | grep -vE 'node_modules|\.git|dist|build|test' | head -8

Instructions

Add health check endpoint(s) to this service per $ARGUMENTS. Match the project's framework and how it registers routes — don't introduce a new server or router.

Method

1. Detect the framework and an existing route to copy registration style, middleware, and conventions (Express/Fastify/Nest, Flask/FastAPI/Django, Gin/Echo, Rails, Spring). Mirror it.
2. Distinguish liveness from readiness — they answer different questions, so prefer two endpoints:

• Liveness (/livez or /health) — is the process up? Cheap, no dependency calls; returns 200 unless the process is broken. Orchestrators restart on failure, so it must not fail just because a dependency is down.
• Readiness (/readyz or /health/ready) — can it serve traffic right now? Checks critical dependencies (db, cache, queue, required downstreams) and returns 503 when any is unavailable so the load balancer stops routing to it.

1. Check real dependencies, cheaply. Use a lightweight probe per dependency (SELECT 1, a Redis PING, a fast head request) with a short timeout so a hung dependency can't hang the check. Run them concurrently where the language makes it easy.
2. Return a useful, safe body. A JSON payload with overall status and per-dependency up/down + latency aids debugging — but don't leak connection strings, credentials, internal hostnames, or versions to an unauthenticated endpoint. Set the right status code (200 healthy / 503 not ready) since probes key on it.
3. Wire it up following the project's routing, and note any config (the path, whether it should bypass auth/rate-limiting, the orchestrator probe settings to point at it).

Rules

1. Liveness must not depend on external services — tying it to the db means a db blip triggers pointless restarts. Keep dependency checks in readiness.
2. Every dependency probe needs a timeout; a health check that can hang is worse than none.
3. Don't leak sensitive detail (secrets, internal topology, exact versions) in the response or expose an internal-only check publicly without auth.
4. Reuse the app's existing db/cache clients and config — don't open new connections per request or hardcode endpoints.
5. After adding, summarize the endpoints, what each checks, the status codes, and the probe configuration to use (e.g. Kubernetes livenessProbe/readinessProbe).

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Pradnyeshp
• Source: Pradnyeshp/Claude-Skills
• License: MIT

Install and usage instructions live in the source repository linked above.