midnightntwrk/midnight-expert/doctor

Expert Doctor

Comprehensive diagnostic and health report for the midnight-expert ecosystem.

Usage

• /midnight-expert:doctor — run diagnostics interactively, offer fixes one at a time
• /midnight-expert:doctor --auto-fix — install missing dependencies silently, prompt only for upgrades and preference choices

Step 1: Launch & Ask (concurrent)

Launch all 5 diagnostic agents in background AND ask the user a question — in a single message with 6 tool calls (5 Agent + 1 AskUserQuestion).

Each agent has subagent_type: "general-purpose" and run_in_background: true. Each agent must run its bash command and return only the raw output lines. Do not include markdown fences or any other text in the agent prompt beyond the instruction.

Agent 1 — Plugin Health

Run the following command and return only the output. No other text.

```

bash "${CLAUDE_SKILL_DIR}/scripts/check-plugins.sh"

```

Agent 2 — MCP Servers

Run the following command and return only the output. No other text.

```

bash "${CLAUDE_SKILL_DIR}/scripts/check-mcp-servers.sh"

```

Agent 3 — External Tools

Run the following command and return only the output. No other text.

```

bash "${CLAUDE_SKILL_DIR}/scripts/check-ext-tools.sh"

```

Agent 4 — Cross-Plugin References

Run the following command and return only the output. No other text.

```

bash "${CLAUDE_SKILL_DIR}/scripts/check-cross-refs.sh"

```

Agent 5 — NPM Registry

Run the following command and return only the output. No other text.

```

bash "${CLAUDE_SKILL_DIR}/scripts/check-npm.sh"

```

AskUserQuestion

Would you also like to check Midnight Tooling status? (Compact CLI, compiler, devnet, proof server)

Step 2: Handle Response

• If the user says yes: invoke the midnight-tooling:doctor skill (via Skill tool) and wait for it and all 5 background agents to complete.
• If the user says no: wait for the 5 background agents to complete.

Step 2.5: Handle Agent Failures

If any background agent fails (returns an error instead of structured output), include a single row in the corresponding report section:

| <section-name> | FAIL | Agent error: <error message> |

Do not skip the section — partial results are better than silence.

Step 3: Present Health Report

Parse all agent output. Each script outputs lines in the format:

CHECK_NAME | STATUS | DETAIL

Map each STATUS to a badge:

• pass → PASS
• warn → WARN
• critical → FAIL
• info → INFO

Present a single formatted report. Omit any section where all checks pass and there is an ALL_*_PASS summary line — only show sections with issues or mixed results. If ALL sections pass, show a brief "all clear" summary.

## Midnight Expert — Health Report
 
### Midnight Tooling (only if user opted in)
(include the delegated report from midnight-tooling:doctor)
 
### Plugins
| Check | Status | Details |
|-------|--------|---------|
| plugin-name | STATUS | details... |
 
### MCP Servers
| Check | Status | Details |
|-------|--------|---------|
| server-name | STATUS | details... |
 
### External Tools
| Check | Status | Details |
|-------|--------|---------|
| tool-name | STATUS | details... |
 
### Cross-Plugin References
| Check | Status | Details |
|-------|--------|---------|
| source → target:ref | STATUS | details... |
 
### NPM Registry
| Check | Status | Details |
|-------|--------|---------|
| check-name | STATUS | details... |

Do not show any intermediate bash output to the user. The report above is the only user-facing output.

Step 4: Offer Fixes

Read references/fix-table.md for the fix recipes.

For each FAIL or WARN item in the report, determine the appropriate fix from the fix-table. Use the platform info line from check-ext-tools.sh to select macOS vs Linux commands.

Platform support: Midnight Expert is developed and tested on macOS and Linux only. If the platform line reports windows (a platform support FAIL is also emitted), the user is on native Windows (Git Bash/MSYS2/Cygwin), which is untested and unsupported. Do not attempt tool-by-tool fixes — instead recommend they run everything inside WSL, per the Platform Support section of references/fix-table.md. WSL reports as linux, so this never fires for WSL users.

If `$ARGUMENTS` contains `--auto-fix`:

• Apply auto-fixable items silently (installs, enables, MCP adds via claude mcp add)
• Always prompt before upgrading outdated tools — show current vs latest version
• Always prompt for MCP server scope (global vs local .mcp.json)
• Log each action taken

If no `--auto-fix`:

• Present each fix one at a time using AskUserQuestion with confirm/skip options
• Group related fixes when possible (e.g., "3 plugins need installing — install all?")

Step 5: Verify & Summary

After applying any fixes, re-run only the scripts whose checks had issues to confirm resolution. Do not re-run passing scripts.

Present a final summary:

### Summary
- FAIL: N
- WARN: N
- PASS: N
- INFO: N
 
[If issues were fixed] Fixed N issue(s) this session.
[If remaining issues] N issue(s) require manual intervention.
[If all green] Midnight Expert ecosystem is healthy and ready for development.

Additional Resources

Scripts

• ${CLAUDE_SKILL_DIR}/scripts/check-plugins.sh — Plugin installation and version checks
• ${CLAUDE_SKILL_DIR}/scripts/check-mcp-servers.sh — MCP server configuration and connectivity
• ${CLAUDE_SKILL_DIR}/scripts/check-ext-tools.sh — External CLI tool availability and versions
• ${CLAUDE_SKILL_DIR}/scripts/check-cross-refs.sh — Cross-plugin skill / agent / command reference validation
• ${CLAUDE_SKILL_DIR}/scripts/check-npm.sh — NPM registry reachability and @midnight-ntwrk scope accessibility

References

• references/fix-table.md — Fix recipes for all issue types with platform-specific commands