Agent error messages expose raw technical details to users

[graycyrus]

Bug

When an agent job fails, the raw technical error message is shown directly in the chat UI:

agent job failed: error sending request for url (https://staging-api.alphahuman.xyz/openai/v1/chat/completions)

Expected behavior

• Technical error details (URLs, stack traces, internal service names) should never be shown to end users in the chat.
• Errors should be reported to Sentry automatically with full context (URL, status code, request ID, etc.).
• The chat should show a user-friendly message like:
  • "Something went wrong. Please try again."
  • "This error has been reported. You can also report it on Discord."
  • Include a "Report on Discord" link/button.

Current behavior

Raw error strings including internal API URLs are displayed verbatim in the chat bubble.

Why this matters

1. UX: Technical jargon is confusing and unhelpful for users.
2. Security: Leaking internal API URLs and infrastructure details is an unnecessary exposure.
3. Observability: Without Sentry capture, these errors are invisible to the team unless a user reports them manually.

Suggested approach

• In the agent execution path, catch errors and:
  1. Log/report to Sentry with full technical context.
  2. Surface a sanitized, user-friendly message in the chat.
• Consider a generic error boundary for all agent job results that strips technical details.

[graycyrus]