doctor-triage

Doctor Triage

Overview

Run diagnostic checks on a Gas Town or beads workspace, interpret each finding, fix what's fixable, and surface unfixable issues that may be upstream bugs worth filing as beads.

When to Use

• User asks to run or fix doctor warnings
• Something in the workspace feels broken or misconfigured
• After upgrades, upstream syncs, or rig additions
• Periodic health checks

When NOT to Use

• Setting up a new rig from scratch (use gastown-rig-setup)
• Debugging a specific test failure (use systematic-debugging)

Step 0: Pre-flight Checks

Verify the required tools are available before proceeding:

gt --version 2>/dev/null
bd --version 2>/dev/null

• If only gt is available → skip beads diagnostics, only offer "Gas Town (gt doctor)"
• If only bd is available → skip Gas Town diagnostics, only offer "Beads (bd doctor)"
• If neither is available → inform the user and exit

Step 1: Ask Which Doctor

Use AskUserQuestion:

"Which workspace do you want to diagnose?"
Options:
  - "Gas Town (gt doctor)" — Town-wide: clones, daemon, Dolt, patrols, sessions
  - "Beads (bd doctor)" — Repo-level: database, schema, hooks, dependencies
  - "Both" — Run gt doctor first, then bd doctor

Step 2: Run Diagnostics

Gas Town

gt doctor -v

Key flags if issues found:

• gt doctor --fix — auto-fix fixable issues
• gt doctor --rig <name> — check specific rig only
• gt doctor --fix --restart-sessions — also restart stale patrol sessions

Beads

Run from the relevant repo root (or mayor/rig/ for Gas Town rigs):

bd doctor -v

Key flags if issues found:

• bd doctor --fix — auto-fix with confirmation
• bd doctor --deep — full graph integrity validation
• bd doctor --server — Dolt server connectivity checks
• bd doctor --check=validate --fix — data-integrity repairs

Step 3: Interpret and Triage

Parse each warning/error into one of three categories:

Category	Action
Auto-fixable	Run --fix and verify resolution
Manually fixable	Investigate root cause, apply targeted fix, verify
Unfixable / upstream bug	Explain to user, collect for Step 5

Triage Guidelines

Do NOT blindly loop --fix. If --fix didn't resolve an issue on the first pass, investigate manually. Looping --fix can cycle without progress.

Do NOT restart the daemon unless the user asks or the diagnostic specifically recommends it.

Known false positives to skip:

• bd doctor reports "No dolt database found" in server mode — this checks for local .beads/dolt/ which doesn't exist with a centralized Dolt server. Verify with bd list instead.

Step 4: Fix What You Can

For each auto-fixable or manually fixable issue:

1. Explain what the issue is and what the fix does
2. Apply the fix
3. Re-run the relevant doctor command to verify resolution
4. Report result to user

After all fixes, run a final clean check:

# Gas Town
gt doctor -v

# Beads
bd doctor -v

Step 5: Reflect and Surface Upstream Bugs

This step is MANDATORY after every doctor run, even if all checks pass or all issues were fixed. Do not skip it.

5a. Reflect on the Full Run

After all fixes are applied and the final clean check is done, review the entire session:

1. Issues that --fix resolved — Were any of these symptoms of deeper bugs? (e.g., fix worked but required workarounds, fix needed multiple passes, fix hint was wrong)
2. Issues fixed manually — Did the manual fix reveal a gap in --fix automation?
3. Issues that --fix failed to resolve — Why? Silent failure? Missing capability?
4. Workarounds applied — Any workaround suggests the tool should handle it natively
5. Misleading or incorrect doctor output — Wrong hints, stale data sources, missing error messages

Build a list of candidate upstream bugs. Include anything that is NOT a configuration problem — i.e., something that would affect any user of gt/bd, not just this workspace.

5b. Offer to Capture Upstream Issues

If NO candidates were identified, briefly state: "No upstream bugs identified — all issues were configuration-related." and skip to the end.

If candidates were identified, explain what was found and offer an escape hatch:

Use AskUserQuestion:

"I identified [N] potential upstream gt/bd issues during this doctor run — things that would affect any user, not just this workspace. Would you like to capture beads documenting these?"
Options:
  - "Yes — show me what you found" — Review and selectively file beads
  - "No — just fix my workspace" — Skip bug filing entirely

If the user declines, stop here. Do not proceed to 5c/5d.

If the user accepts, present the specific candidates:

Use AskUserQuestion with multiSelect:

"Which issues should we file?"
Options (multiSelect):
  - "#1 [short title]" — [what happened, why it's a bug, not a config issue]
  - "#2 [short title]" — [what happened, why it's a bug, not a config issue]
  - ...

If no issues are selected, stop here.

5c. Ask Where to File

The user may have different rigs set up for tracking bugs. Do NOT assume a specific rig.

Use AskUserQuestion:

"Which rig should these bugs be filed in?"
Options:
  - List each rig from `gt rig list` output
  - Include a description hint: for a rig named "gastown", say "Gas Town codebase (gt bugs)"
  - For a rig named "beads", say "Beads codebase (bd bugs)"

If bugs span both gt and bd codebases, ask once and let the user choose — or offer "Split by codebase" to file gt bugs in one rig and bd bugs in another.

5d. Create Beads

For each selected bug, create a bead using bd q (quick capture):

bd q -p P2 -l bug -l gastown -l "gt-doctor" "<title>"
# or for beads bugs:
# bd q -p P2 -l bug -l beads -l "bd-doctor" "<title>"

Then add a detailed description with bd update:

bd update <id> --description "<description>"

Description template:

## Bug

[What went wrong — observable behavior]

## Expected Behavior

[What should have happened]

## Actual Behavior

[What actually happened, including exact error output or silent failure]

## Reproduction

1. [Step-by-step to reproduce]

## Workaround

[If one was applied during this session]

## Environment

- gt version: [from `gt --version` or stale-binary check output]
- bd version: [from `bd --version` or beads-binary check output]
- Platform: [OS]

Labels:

• bug — always
• gastown or beads — based on which codebase the bug is in
• gt-doctor or bd-doctor — to group doctor-specific bugs

Priority:

• P1 — blocks normal operation (e.g., data loss, can't start daemon)
• P2 — wrong behavior but has workaround (most doctor bugs)
• P3 — cosmetic or hint text issues

Quick Reference

Tool	Basic	Verbose	Fix	Deep
gt doctor	gt doctor	gt doctor -v	gt doctor --fix	N/A
bd doctor	bd doctor	bd doctor -v	bd doctor --fix	bd doctor --deep

Scenario	Command
After upstream sync	gt doctor -v then bd doctor -v
After rig addition	gt doctor --rig <name> then bd doctor -v (from rig)
Dolt connectivity	bd doctor --server
Data integrity	bd doctor --check=validate --fix
Stale sessions	gt doctor --fix --restart-sessions

Common Mistakes

Mistake	Fix
Looping --fix when it doesn't resolve	Investigate manually after first --fix attempt
Running bd doctor from wrong directory	Must be in repo root or mayor/rig/ for Gas Town rigs
Treating server-mode "No dolt database" as real	Use bd list to verify — it's a false positive
Restarting daemon for every issue	Only restart when diagnostic says to or user asks
Using gt up --restart for stale settings	gt up --restart doesn't exist. Use gt doctor --fix --restart-sessions to delete stale Claude settings and recreate them properly