Dirac Debug Ops Skill
Use When
- User reports "cannot execute", "no state sync", or "only trigger text seen".
- Need one command to capture execution bus truth and Octopus runtime evidence.
- Need controlled dispatch test with before/after snapshots.
Primary Tool
scripts/run_dirac_debug_ops.py
Typical Commands
# Snapshot only (non-intrusive)
python scripts/run_dirac_debug_ops.py --mode snapshot
# Dispatch only (route + execute)
python scripts/run_dirac_debug_ops.py --mode dispatch --task "Dirac_solver 调试" --execute --auto-replan
# Full check: snapshot -> dispatch -> snapshot
python scripts/run_dirac_debug_ops.py --mode full --task "Dirac_solver 调试 /auto smoke" --execute --auto-replan
Required Outputs
- Worker status (
pid,alive). - Queue depth and latest task status.
- Bridge
execution_bus.last_tasksnapshot. - Progress sync (
state/dirac_solver_progress_sync.json) phase/checks. - Octopus evidence fields when available:
- SCF convergence and iterations
- total energy
- HOMO/LUMO and gap
- TD energy trajectory (
step,time,total_energy)
Rules
- Prefer
--mode snapshotfirst when diagnosing production issues. - For dispatch tests, include both before and after snapshots.
- If dispatch is blocked, return exact
dispatch_statusandfailure_reasonfrom tool output. - Never conclude "no execution capability" without queue + bridge + sync evidence.