Steps

Read the conflicts file:

cat ~/.claude/brain-conflicts.json 2>/dev/null || echo '{"conflicts":[]}'

Filter to unresolved conflicts (where resolved is not true).
If no unresolved conflicts, tell the user: "No pending conflicts. Brain is fully synced."
For each unresolved conflict, present:
- The topic
- What Machine A says
- What Machine B says
- The AI's suggestion and confidence level
Ask the user to choose:
- Accept AI suggestion: Apply the suggestion
- Keep Machine A's version: Use A's content
- Keep Machine B's version: Use B's content
- Keep both (machine-specific): Tag each with its machine name
- Custom: Let the user type their own resolution
After each resolution:
- Mark the conflict as resolved: true with the chosen resolution in brain-conflicts.json
- Apply the resolution to the appropriate brain file (CLAUDE.md, memory, etc.)

After all conflicts are resolved:

bash "${CLAUDE_PLUGIN_ROOT}/scripts/push.sh"

brain-conflicts