t150: Supervisor self-healing: auto-create diagnostic subtask on failure

[marcusquinn]

When a task fails or gets blocked, the supervisor should:

1. Analyze the failure error message
2. Create a diagnostic subtask (e.g., t147.3.diag) with a prompt that includes:
   • The original task description
   • The error message
   • The last N lines of the worker log
   • Instructions to diagnose root cause and either fix the issue or create a targeted fix task
3. Add the diagnostic subtask to the same batch
4. The diagnostic worker can then either:
   • Fix the root cause and re-queue the original task
   • Create a more targeted subtask if the fix is complex

This creates a self-improving feedback loop where failures are automatically investigated
and resolved without human intervention during long-running sessions.

Related: t149 (auto-create GH issues), t147.6 (adaptive concurrency)

[marcusquinn]

Already completed. t150 implemented in PR #462 (merged 2026-02-07). Self-healing auto-creates diagnostic subtasks on failure/block.