name	railform-railway-stack
description	Deploy, preview, approve, and troubleshoot Railway infrastructure through Railform. Use this whenever the user mentions Railform, railform.config.ts, railform plan/apply, agent-safe Railway changes, approval handoffs, or Railway services/databases managed declaratively.

Railform Railway Stack

Start here when creating, previewing, deploying, modifying, or troubleshooting Railway infrastructure through Railform.

Load Only What You Need

Read railform.md for command examples, config APIs, and common gotchas.
Keep all infrastructure changes in railform.config.ts or in Railform itself.
If Railform is missing a capability needed for the deployment, improve Railform instead of reaching for another infrastructure tool.

Default Agent Flow

Discover the application before writing config: package manager, start command, required env vars, health route, database usage, git source, and whether it serves HTTP.
Run railform --help, then railform plan. If config loading fails because @railform/core is missing, install it with the project package manager and rerun railform plan.
Edit railform.config.ts until the plan matches the intended Railway stack. Infer app service source.repo from git remote -v and source.branch from the current branch when the repo is GitHub-backed.
For HTTP services with a health check, ask or default to a public domain. If Railform cannot declare that domain yet, say so explicitly before apply.
Request human review with railform apply --request-approval --format json.
Tell the human to run the returned railform review and railform approve commands.
Continue with railform apply --approval <approval-id> --wait --format json.
Treat status: applied as config commit success, not deployment completion. Verify Railway deploy health, domain availability, or a successful HTTP health check before calling the deployment complete.
If apply reports a service, deploy, image, or configuration failure, surface that exact failure to the user and update the config or Railform code before trying again.

Agent Safety

Do not run plain railform apply unless the user explicitly asked for direct mutation.
Do not wait on prompts. Pass prompt values with --var SERVICE.KEY=value or --var KEY=value.
Use randomSecret(length) for generated secrets and ask the human only for external secrets that Railform cannot generate.
Use RAILFORM_DANGEROUSLY_SKIP_PERMISSIONS=1 only when the user explicitly asks to bypass the approval handoff.
Report failures with concrete resource names, approval IDs, deployment IDs, statuses, and relevant error text.
In recording or demo contexts, do not relay workspace-list output. Prefer an existing RAILWAY_WORKSPACE_ID, saved .railform/state.json, or a user-provided workspace name; if a command enumerates workspaces anyway, summarize only the selected workspace.
Remove generated placeholder variables like API_TOKEN, SESSION_SECRET, or INTERNAL_API_URL unless the inspected app actually needs them.

Railform Railway Stack

Start here when creating, previewing, deploying, modifying, or troubleshooting Railway infrastructure through Railform.

Load Only What You Need

Read railform.md for command examples, config APIs, and common gotchas.

Keep all infrastructure changes in railform.config.ts or in Railform itself.

If Railform is missing a capability needed for the deployment, improve Railform instead of reaching for another infrastructure tool.

Default Agent Flow

Discover the application before writing config: package manager, start command, required env vars, health route, database usage, git source, and whether it serves HTTP.

Run railform --help, then railform plan. If config loading fails because @railform/core is missing, install it with the project package manager and rerun railform plan.

Edit railform.config.ts until the plan matches the intended Railway stack. Infer app service source.repo from git remote -v and source.branch from the current branch when the repo is GitHub-backed.

For HTTP services with a health check, ask or default to a public domain. If Railform cannot declare that domain yet, say so explicitly before apply.

Request human review with railform apply --request-approval --format json.

Tell the human to run the returned railform review and railform approve commands.

Continue with railform apply --approval <approval-id> --wait --format json.

Treat status: applied as config commit success, not deployment completion. Verify Railway deploy health, domain availability, or a successful HTTP health check before calling the deployment complete.

If apply reports a service, deploy, image, or configuration failure, surface that exact failure to the user and update the config or Railform code before trying again.

Agent Safety

Do not run plain railform apply unless the user explicitly asked for direct mutation.

Do not wait on prompts. Pass prompt values with --var SERVICE.KEY=value or --var KEY=value.

Use randomSecret(length) for generated secrets and ask the human only for external secrets that Railform cannot generate.

Use RAILFORM_DANGEROUSLY_SKIP_PERMISSIONS=1 only when the user explicitly asks to bypass the approval handoff.

Report failures with concrete resource names, approval IDs, deployment IDs, statuses, and relevant error text.

In recording or demo contexts, do not relay workspace-list output. Prefer an existing RAILWAY_WORKSPACE_ID, saved .railform/state.json, or a user-provided workspace name; if a command enumerates workspaces anyway, summarize only the selected workspace.

Remove generated placeholder variables like API_TOKEN, SESSION_SECRET, or INTERNAL_API_URL unless the inspected app actually needs them.

railform-railway-stack