Development Environment

The codecollection-devtools container is the standard development environment for authoring and testing Skill Templates (CodeBundles). One image works with any codecollection — set an env var and start developing.

Three ways to run

Method	Best for	Setup
GitHub Codespaces	PR review, zero local setup	Create a codespace from the devtools repo
VS Code devcontainer	Local development with Docker	Clone devtools repo, reopen in container
`docker run`	Headless / CI	`docker run -it ghcr.io/runwhen-contrib/codecollection-devtools:latest`

All three pull the pre-built image from GHCR — no local Docker build required.

Bootstrapping a codecollection

Inside the container, run:

# defaults to rw-cli-codecollection on main
task setup

# specific repo + PR
task setup REPO=runwhen-contrib/rw-cli-codecollection PR=42

# different codecollection and branch
task setup REPO=runwhen-contrib/azure-c7n-codecollection BRANCH=feat/foo

task setup clones the repo, installs Python deps, installs authoring skills as Cursor rules, and checks out the PR branch if specified.

The `ro` runner

ro is a thin wrapper around robot that provides:

Isolated working directories — each run gets its own temp dir with copied cloud CLI configs (.azure, .gcloud, .kube)
Log output — HTML reports written to /robot_logs, served at localhost:3000
Selective test execution — ro --test "Check Health" runbook.robot

ro                              # run all .robot files in current dir
ro runbook.robot                # run a specific file
ro --test "Check Health"        # run a specific test case
ro ../other-codebundle/         # run tests in a different directory

Credentials

Mount or copy credentials into the auth/ directory:

/home/runwhen/
├── auth/
│   ├── .kube/config        # kubectl
│   ├── .azure/             # Azure CLI
│   └── .gcloud/             # Google Cloud SDK
├── codecollection/
│   ├── codebundles/        # your Skill Templates
│   └── libraries/          # shared keyword libraries
└── ro                      # test runner

ro copies these configs into an isolated temp directory per run, so parallel executions don’t interfere with each other.

Authoring skills

The skills/ directory in the devtools repo contains platform-specific authoring guidance that is automatically installed as Cursor rules during task setup. These give AI assistants (and human authors) context about generation rules, SLI patterns, and test infrastructure conventions.

Skill	Covers
`generation-rules-kubernetes.md`	K8s resource types, match rules, qualifiers, templates
`generation-rules-aws.md`	AWS resource types, CloudQuery tables, account qualifiers
`generation-rules-azure.md`	Azure + Azure DevOps platforms, resource groups, subscriptions
`generation-rules-gcp.md`	GCP resource types, project qualifiers
`sli-authoring.md`	In-repo SLIs, cron-scheduler SLIs, scoring patterns
`test-infra-kubernetes.md`	Static manifests, Terraform patterns, Taskfile contract
`test-infra-azure.md`	Azure Terraform test infra, tf.secret, workspaceInfo
`test-infra-azure-devops.md`	DevOps projects, pipelines, agent pools via Terraform
`test-infra-cloud.md`	Shared conventions across all cloud platforms
`skill-template-authoring.md`	General Skill Template structure and conventions

Re-install after an update:

task install-skills

Dev mode (`RW_MODE=dev`)

The devtools container sets RW_MODE=dev, which activates rw-core-keywords dev mode. In dev mode:

Secrets are read from environment variables or local files (not Vault)
Metrics are logged to console (not pushed to OpenTelemetry)
Issues are logged to console (not written to issues.jsonl)
Reports are logged to console (not written to report.jsonl)

Use RW_SECRET_REMAP to map variable names to different env var keys, and RW_FROM_FILE to read secret values from local files:

export RW_MODE=dev
export RW_FROM_FILE='{"kubeconfig":"/home/user/auth/kubeconfig"}'
export RW_SECRET_REMAP='{"my_token":"MY_LOCAL_TOKEN_VAR"}'

See the RW.Core Keyword Reference for the full dev-mode API, and the Runtime Stack section for how dev mode differs from production.