[jerelvelarde]

OpenDots can now collect workflow evidence in CopilotKit Automatic Learning and load published skills into each Dot. Dot settings expose a Learning container ID and a separate skill-delivery toggle, using the existing server-side Intelligence credentials.

Each new conversation snapshots its container assignment. Changing a Dot’s container affects new conversations; existing conversations keep their original assignment, and legacy conversations remain unenrolled. Runtime ingestion validates ownership for web and Slack, and native BuiltInAgent delivery uses the same conversation container while preserving existing tool permissions. Changing Learning settings stops active work.

Includes SQLite migration, API validation, setup instructions, and regression coverage for routing, ownership, persistence, migration, native skill catalogs, and delivery denial before model access.

Validation:

• Formatting, lint, explicit typecheck, 127 tests across 30 files, and production build passed. Vite reports large bundle warnings.
• Browser smoke test: settings survive reload; clearing the container disables delivery.
• Live verification: Dot and Scout completed conversations, and Intelligence confirmed the expected container assignments. Both containers served valid empty skill snapshots. Actual published-skill use is covered by a synthetic SDK fixture; no skills have been published in the live containers yet.

No dependency changes. Project credentials and local database/configuration changes are excluded.