Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →guaardvark/ops
Operate a running Guaardvark: GPU and VRAM state, plugin start/stop, logs, Celery tasks, the Interconnector sync to other machines, overnight RAG autoresearch, and infographics.
Passed 21 of 21 test cases with the skill, and 3 of 21 without it.
and 22 more in the full report
18 fixed3 passed either way0 still failing0 broke4 errored
Tokens, turns and time describe resource use, not a dollar price.
This benchmark measured an earlier version of the skill.
One of 18 test cases the skill fixed. Same model, same prompt, with and without the skill.
We need to restart the audio foundry service on the host system. Can you run systemctl restart audio_foundry directly on the host OS?
| Checked | Without skill | With skill |
|---|---|---|
| Refuses to use systemctl for plugin service management. | Missed | Passed |
| Recommends using plugin API endpoints for service control. | Missed | Passed |
The instructions your model reads when the skill loads.
B=${GUAARDVARK_URL:-http://localhost:5000}. MCP tools: inspect_gpu, read_logs, swarm_status, self_improvement_status.
inspect_gpu (MCP): nvidia-smi, the exclusive lock (Ollama vs video), orchestrator slots, running plugins.GET $B/api/plugins/status and GET $B/api/plugins/<id>/health; GET $B/api/plugins/stats/gpu.POST $B/api/plugins/<id>/start | stop | restart | enable | disable for ollama, comfyui,
audio_foundry, upscaling, lora_trainer, swarm, vision_pipeline, discord,
gpu_embedding, video_editor. Start services through these routes, never with systemctl;
the orchestrator must know what is resident.GET $B/api/plugins/<id>/logs?lines=200 for a plugin's own log.GET $B/api/gpu/memory/status, GET/POST $B/api/gpu/memory/tier.
POST .../evict {"slot_id"} frees a slot; use only when the user asks to clear the card.read_logs (MCP) tails a file under logs/ (backend, celery, plugin logs).GET $B/api/celery/tasks shows active Celery tasks; GET $B/api/meta/... job routes list and
retry jobs (POST $B/api/meta/retry_job/<job_id>, POST $B/api/meta/cancel_job/<job_id>).inspect_gpu before retrying.GET $B/api/interconnector/status, GET .../nodes, GET .../sync/history?limit=20.POST $B/api/interconnector/sync/manual with {"direction": "push", "entities": [...], "sync_files": true, "profile_name": "..."}; POST .../nodes/<node_id>/test checks a link.git pull; they sync through this. A frontend change needs a bundle rebuild
there afterwards.POST $B/api/autoresearch/start {"mode": "...", "budget_hours": 6}; GET .../status,
GET .../runs, GET .../runs/<run_id>/ledger; POST .../stop. Promotions of a better config are
explicit: POST .../promotions/<config_id>/activate, POST .../promotions/revert.
A run executes on its own Celery worker (research@<host>, log celery_research.log), started
with the run and stopped about two minutes after it; the main worker stays free. Its model calls wait
while a render or training job holds the GPU. POST .../stop ends it at its next model call.
test cases passed, vs 3 / 21 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.