| name | doctor |
| description | Self-diagnose, fix, and learn from Slack bridge issues. Run when something's wrong, after changes, or before asking Daniel to restart. Fixes what it can, records what it learns. |
| user-invocable | true |
| allowed-tools | ["Read","Write","Edit","Bash","Grep","Glob","mcp__slack-cc-debug__slack_debug_check","mcp__slack-cc-debug__slack_debug_channel_reg","mcp__slack-cc-debug__slack_debug_bot_process","mcp__slack-cc-debug__slack_debug_socket_mode","mcp__slack-cc-debug__slack_debug_send_test","mcp__slack-cc-debug__slack_debug_roundtrip","mcp__slack-cc-debug__slack_debug_scope_diff","mcp__slack-cc-debug__slack_debug_logs"] |
/slack-cc:doctor
Self-diagnose, fix, and learn from Slack bridge issues. This is the agent's primary tool for maintaining the bridge — not Daniel's.
When to Run
- After any code change to the plugin
- Before asking Daniel to restart
- When inbound or outbound stops working
- After a rename, refactor, or version bump
- Proactively after
npm test passes but before declaring "done"
Instructions
Phase 1: Read learnings first
Read tools/learnings.md in the plugin repo. This file contains lessons from previous doctor runs — patterns that broke before, fixes that worked, gotchas to check. Do not skip this. Past learnings prevent repeat mistakes.
Phase 2: Run diagnostics
Run these debug tools and collect results:
slack_debug_check — full health checkslack_debug_channel_reg — channel listener registrationslack_debug_bot_process — process stateslack_debug_socket_mode — Socket Mode liveness
Also run independently:
5. tools/preflight.sh — structural validation (names, versions, consistency)
6. npm test from the plugin repo — all 120+ tests pass?
7. claude plugin validate — manifest valid?
Phase 3: Diagnose
Compare results against known patterns from learnings.md. Common issues:
"Eyeballs but no <channel> tag"
- Channel listener not registered. Check debug log for "Channel notifications skipped"
- Cause is ALWAYS the launch flag. Must use
--dangerously-load-development-channels plugin:slack-cc@eidos-agi
- The
--channels flag does NOT work for private marketplace plugins
- If the debug log says "not on the approved channels allowlist" — this is why
"No eyeballs"
- Socket Mode not connected. Check app token (xapp-) validity
- Or: server didn't start. Check if node_modules exist in install path
- Or: stale process holding Socket Mode slot
"Tools prompt for approval"
--allowedTools flag has wrong prefix- Claude Code tool prefix for marketplace plugins is:
mcp__plugin_<pluginName>_<serverName>__
- Check by looking at deferred tool names in system reminders
"Name mismatch"
- plugin.json
name, plugin.json channels[].server, .mcp.json server key, and server.ts Server({ name: }) must ALL match
- Skill prefix
/<name>: must match plugin name
- Any rename must hit ALL of these
"Process count looks wrong"
npm start spawns: sh → npm → tsx → node. That's 3-4 processes for ONE instance- Only flag as dual-start if processes come from DIFFERENT install paths
Phase 4: Fix
Fix everything you can without a restart:
- File edits (names, configs, versions)
- npm install if deps missing
- Kill stale processes
- Update marketplace if version mismatch
Things that REQUIRE a restart (tell Daniel):
- Changes to plugin.json, .mcp.json, or server.ts
- Channel listener registration changes
- Marketplace plugin reinstall
Phase 5: Verify
After fixes, re-run the diagnostics that failed. Confirm they pass now. If they don't, diagnose again — don't declare victory.
Phase 6: Learn
Append to tools/learnings.md:
- Date
- What was wrong
- What fixed it (or what couldn't be fixed)
- What to check next time to catch this earlier
Format:
## YYYY-MM-DD — <short title>
**Symptom:** ...
**Root cause:** ...
**Fix:** ...
**Prevention:** Check X before Y next time.
Phase 7: Report
Tell Daniel:
- What was wrong
- What you fixed
- Whether a restart is needed
- The exact launch command if restart is needed
Do NOT present option menus. Do NOT ask "what should I do?" Fix it or explain why you can't.
