Autter Skills

AI agent skills that set up Autter for you — wire Autter Runtime — open-source error tracking, usage telemetry, and LLM tracing — into any codebase, regardless of language or framework.

Drop these into Claude Code, Cursor, Codex, or any editor that supports the Agent Skills standard and start using them immediately.

How to use this

npx skills add Autter-dev/autter-skills --all

Then tell your agent what you want:

• "use the skills to install Autter Runtime in this project." — runs autter-runtime-setup, which inventories your repo and routes each service to the right style skill automatically — you don't need to pick one yourself.

Want just one skill? npx skills add Autter-dev/autter-skills --skill otel-node-style

What's inside

Skill	Covers
autter-runtime-setup	Start here. Inventories the repo (including which services call LLM APIs), gets an ingest key set up, routes each service to the right style skill below, verifies with a key preflight plus a temporary selftest path that proves every wired pipeline — traces/errors, metrics, and (where wired) a fake LLM test trace — per service, then records the error + info instrumentation convention in the repo's agent-instruction files (CLAUDE.md / AGENTS.md / Cursor / Copilot) so new code stays instrumented.
otel-node-style	Node.js (Express, Fastify, Koa, NestJS) and Next.js, via @autter/runtime-node / @autter/runtime-next.
otel-browser-style	Browser apps — React, Vue, Svelte, Angular, vanilla SPA, static sites — via @autter/runtime-browser, including CSP violations and privacy-conscious recent action context.
otel-python-style	FastAPI, Flask, Django, plain WSGI/ASGI, via the standard OpenTelemetry Python SDK.
otel-go-rust-style	Go and Rust backends, via each language's official OTel SDK.
otel-generic-style	Everything else — Java, .NET, PHP, Ruby, Elixir, Kotlin, … — via standard OTLP/HTTP + OTel env vars.

Why this works for any language

Autter Runtime's ingester uses two key types and these HTTP endpoints:

Credential	Lives in	Can
Server key autter_rt_…	backend env vars	send OTLP traces/metrics, relay browser events
Client key autter_rtc_…	frontend bundles (publishable)	send browser events only, origin-restricted

Endpoint	Format
POST /v1/traces, POST /v1/metrics	OTLP/HTTP — protobuf or JSON
POST /v1/browser	compact JSON (@autter/runtime-browser payload)
POST /v1/profiles	symbolized pprof (server key only)
POST /v1/sourcemaps	release-keyed source map JSON (server key only)
POST /v1/platform-events	ECS/Kubernetes OOM and restart JSON (server key only)

Any language with an OpenTelemetry SDK can send server telemetry — that's every mainstream language. Only Node.js and the browser get dedicated first-party npm packages (@autter/runtime-node, @autter/runtime-browser, @autter/runtime-next); everything else is a thin style guide over the standard OTel SDK for that language, which otel-generic-style covers even when no dedicated skill exists yet.

Memory pressure uses that same OTLP metric endpoint in every backend language. The Node package exports process metrics automatically; other stacks enable a process meter or add the portable RSS/heap gauges described in Runtime's memory contract. The feature also needs the 1.3.3+ ingester and Autter backend/frontend deployment. Applications must be redeployed after instrumentation changes; OOM/restart correlation requires an ECS/Kubernetes event forwarder.

Getting an ingest key

Create one on your Autter dashboard: Settings → Access Tokens → Runtime ingest keys → Create key. Pick the repository, then choose:

• Server — secret, for backends and relays. Keep it in env vars, never commit it.
• Client — publishable, for browser-only apps with no backend. Scoped to the exact origins you register it for.

Set the key in your own environment (shell profile, gitignored .env, or secret manager) as AUTTER_RUNTIME_KEY — don't paste key values into the agent chat. The skills are written to only ever reference the env var by name; the value never needs to reach the agent.

Endpoint regressions

The setup skills cover request histograms, release identity, and trace comparison requirements. Node/Next.js examples use SDK 1.3.2 and opt into bounded slow-request retention. External OTel exporters need explicit-bucket delta histograms. Self-hosted ingesters require 1.3.1 or later.

Enabling detection in the platform does not update customer SDKs. Incident feedback lets a team mark a diagnosis as correct, incorrect, or expected. Fixes remain draft pull requests for human review. Production verification uses existing telemetry; selftests belong only in isolated test environments.

Full docs: github.com/Autter-dev/autter-runtime.

License

MIT

Connect existing provider logs

autter-runtime-setup also supports repository-scoped Sentry, PostHog, Grafana/Loki, Datadog and webhook sources. Ask the agent to connect the chosen source in repository Runtime settings and verify stored logs, RCA and eligible draft fixes. Connector-only setup does not require an application SDK or a Runtime ingest key. See external-source guidance.