Local Model Queue Enforcement

Policy: all local-model invocations must route through:

  • skills/ollama-agent-queue/scripts/integration_helper.py
  • or skills/ollama-agent-queue/scripts/queue_manager.py

Runtime guard (single command)

python3 scripts/local_queue_guard.py \
  --context "subagent" \
  --command "<command to execute>"
  • Exit 0: allowed
  • Exit 42: blocked (direct local model invocation detected)

Preflight audit (codebase)

python3 scripts/audit_local_model_routing.py
python3 scripts/audit_local_model_routing.py --json

Latest report path:

  • generated/local_model_routing_audit_2026-02-28.json

Launch-path wiring (implemented)

scripts/spawn_smart.sh now enforces the guard before spawning subagents:

  • pre-checks the task payload with local_queue_guard.py
  • blocks with exit 42 on policy violation
  • surfaces a clean operator-facing error message
  • writes structured violation events to generated/queue_guard_violations.jsonl