Bootstrap Data Seeding

bootstrap-data-seeding.yml is a one-time, post-deploy step that seeds the default configuration the application needs to be usable. You run it once after the first full deploy, and again only when the default catalog changes.

What gets seeded

The workflow runs scripts/stack-bootstrap/seed.sh, which resolves the target table names from SSM and runs backend/scripts/seed_bootstrap_data.py to upsert:

• Default models — the AI models available out of the box.
• RBAC roles — the baseline role definitions.
• Quota tiers — default usage limits.
• Tool catalog — the built-in tool entries.
• Skills — two admin-catalog skills. Web Research Assistant is granted to the default role. Skill Creator (helps users write their own skills, adapted from Anthropic’s Apache-2.0 skill-creator) is granted to no role: grant it to a pilot role under Admin → Roles when you are ready. It saves skills directly for users who also have the Skill Authoring tool (SKILL_AUTHORING_ENABLED), and hands everyone else a draft to add under Customize → Skills.

Safe to re-run

Seeding is idempotent — re-runs upsert rather than duplicate, so it’s safe to run again any time the default catalog changes (a new default tool, model, or role). Re-running won’t disturb data an admin has since customized through the dashboard.

Authentication is not seeded

There’s no auth-provider seeding. Cognito’s first-boot flow handles initial access: the first person to open the application creates the admin account directly, and that account automatically becomes the system admin. Federated identity providers (Entra ID, Okta, Google) are added later from the admin dashboard — no redeploy or seed required.

When to re-run

You changed	Re-run
Default tools, models, roles, or seeded skills	bootstrap-data-seeding.yml

Routine application and infrastructure changes don’t need a re-seed.