Rai Doctor

Doctor

Purpose

Diagnose RaiSE setup health, explain issues in plain language, and guide the user through resolution. The user never touches the CLI directly — Rai runs the diagnostics and translates results into actionable conversation.

Context

When to use: User reports something isn't working, after rai init, after upgrading raise-cli, or proactively at session start when setup feels off.

When to skip: The issue is clearly in user code, not RaiSE configuration.

Steps

Step 1: Run diagnostics

rai doctor --json

Parse the JSON output. Categorize results by severity.

Step 2: Report to user

If all checks pass:

Your RaiSE setup is healthy. All checks passed.

If there are warnings or errors, explain each one conversationally:

• Translate technical check IDs into human language
• Group by category (environment, project, adapters, skills, MCP)
• For each issue: what it means, why it matters, and what to do

Example:

I found 2 issues with your setup:

1. Graph is outdated — your knowledge graph hasn't been rebuilt since governance files changed. This means my context queries may return stale data. I can rebuild it now if you'd like.

2. MCP server 'jira' is unreachable — the Jira integration can't connect. Check that JIRA_URL and JIRA_API_TOKEN are set in your environment.

Step 3: Offer fixes

For auto-fixable issues (those with fix_id in the JSON):

I can fix issue #1 automatically. Should I run rai doctor --fix?

Wait for user confirmation before running fixes.

rai doctor --fix

Report the outcome of each fix.

Step 4: Guide manual fixes

For issues that can't be auto-fixed, guide step by step:

• Missing env vars → tell the user exactly which vars to set and where
• Missing optional extras → provide the exact pip install command
• Stale skills → offer to run rai skill sync
• Invalid manifest → explain what's wrong and how to fix it

Step 5: Bug report (if unresolvable)

If issues persist after fixes:

This looks like a bug in RaiSE itself. I can generate a diagnostic report for the team. It only includes non-sensitive data (versions, check results, file names — never secrets or code). Want me to prepare it?

If the user agrees:

rai doctor report

Show the user the saved report path so they can review it. Then:

The report is saved at {path}. Review it, and when ready I can open your email client to send it:

rai doctor report --send

Output

Quality Checklist

• Never run --fix without user confirmation
• Explain issues in plain language, not technical jargon
• Never expose secrets, tokens, or file contents in conversation
• Offer bug report only after fix attempts fail
• Always show the report path so user can review before sending