Runtime Architecture

Every RunWhen Skill Template runs inside a container built from the same layered stack. Understanding the layers helps you know what’s available at runtime and how dev mode differs from production.

The layers

The runtime image stack: rw-base-runtime provides Python, the worker binary, cloud CLIs, and rw-core-keywords; a CodeCollection image is built from it and adds your codebundles. In production the platform runs persistent worker pods; in development you use the devtools image with RW_MODE=dev.

Layer	Repo	What it provides
rw-base-runtime	rw-base-runtime	Python, the worker binary, all cloud CLIs, `rw-core-keywords` installed system-wide, robot-runtime helper scripts
rw-core-keywords	rw-core-keywords	`RW.Core` — the Robot Framework keyword library for secrets, metrics, issues, reports, services
CodeCollection image	e.g. rw-cli-codecollection	`FROM rw-base-runtime`, adds codebundles + libraries + codecollection-specific pip deps

Dev vs production

	Dev	Production
Image	codecollection-devtools	CodeCollection image (FROM rw-base-runtime)
Mode	`RW_MODE=dev`	`RW_MODE=production` (default)
Secrets	Env vars / local files	Vault, Kubernetes secrets, cloud providers
Metrics	Logged to console	Pushed to OpenTelemetry Collector
Issues	Logged to console	Written to `issues.jsonl`
Reports	Logged to console	Written to `report.jsonl`

Both modes use the same RW.Core keyword library — the behavior difference is handled internally by rw-core-keywords based on the RW_MODE environment variable. Your .robot code doesn’t change between dev and production.