AI-powered "Suggest Next Steps" action for Goals

Learning Companion - AI-powered suggest next steps

Tasks

• Add a "Suggest next steps" action on the goal detail page.
• Send the goal (title, description, status) plus its past LearningSessions to the OpenAI Chat Completions API.
• Prompt for 2-3 concrete, actionable next learning steps.
• Render the reply as a short list on the goal detail page.
• Reuse the existing OpenAI API key configuration from the "Generate summary" ticket (.env / application.properties / Rails credentials) — do not introduce a second config path.

Acceptance criteria

• Clicking "Suggest next steps" on a goal with session history returns 2-3 concrete, relevant suggestions rendered as a list (not raw JSON or a single paragraph).
• Works independently from the "Generate summary" action (doesn't require it to have been run first).
• Handles a goal with no sessions yet (sensible suggestions or a clear empty-state message, no crash).
• Handles OpenAI API errors gracefully (timeout, rate limit, invalid key) with a user-facing error message.
• Tests cover: request payload construction (goal + sessions correctly gathered and scoped to current user), response parsing into a list, and graceful handling of API failure (mock the OpenAI call — do not hit the real API in tests).

Notes for agent

• This depends on Goals/Sessions (#3) for the data, and assumes the OpenAI client/config set up in the "Generate summary" ticket (#5) already exists — reuse it, don't duplicate.
• Mock/stub the OpenAI API call in tests — do not make real API calls during the test suite.
• Parse the model's reply into a list structure before rendering (e.g. split on numbered lines) rather than dumping raw text.

Start at the goal detail page and the existing “Generate summary” OpenAI client/config, reusing the path in .env, application.properties, or Rails credentials. Trace Goals and LearningSessions to ensure data is gathered for the current user, then add mocked tests for payload construction, response-list parsing, and API failures. Done means the action handles session history, empty sessions, and errors without real API calls.