Menu
Mandatory protocol for creating auditable experiments via sweep/run endpoints in Wild V2
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
System prompt for handling experiment alerts. Provides diagnosis guidance, GPU wrapper context, action suggestions, and structured response from allowed choices.
Default system prompt for agent chat mode. Provides identity, environment context, compute awareness, API-driven job submission, and workflow reflection.
Generates a structured experiment plan with compute-aware recommendations and saves it via the plan API endpoint.
Wraps user steering input with context signals for the model during a wild loop session
Ports new models into FastVideo with strict numerical alignment to official implementations. Use when adding a FastVideo model/pipeline, porting an official or Diffusers checkpoint, or debugging parity/alignment.
Single source of truth protocol for Wild V2 preflight, sweep/run auditability, GPU discovery, and parallel scheduling
| name | wild_v2_sweep_run_audit_protocol |
| description | Mandatory protocol for creating auditable experiments via sweep/run endpoints in Wild V2 |
| category | protocol |
| variables | [] |
This protocol is mandatory for experiment execution in Wild V2.
Only runs created through server sweep/run endpoints are visible to users in the UI and audit logs.
If you execute experiments directly in shell (for example python train.py ..., bash run.sh, torchrun ...) without creating runs through the API:
Treat local direct execution as non-compliant.
chat_session_id in all sweep/run creation request bodies.curl -X POST "$SERVER_URL/sweeps/wild" \
-H "Content-Type: application/json" \
-H "X-Auth-Token: $AUTH_TOKEN" \
-d '{
"name": "mnist-mlp-cpu-grid",
"goal": "one-layer MLP grid search on MNIST CPU",
"chat_session_id": "$CHAT_SESSION_ID"
}'
Extract id from response as SWEEP_ID.
curl -X POST "$SERVER_URL/runs" \
-H "Content-Type: application/json" \
-H "X-Auth-Token: $AUTH_TOKEN" \
-d "{
\"name\": \"mnist-mlp-lr1e-2-bs128-seed1\",
\"command\": \"cd /path/to/workdir && python scripts/train_mnist_mlp.py --lr 1e-2 --batch-size 128 --seed 1 --device cpu --log-dir exp/logs --out-dir exp/outputs\",
\"sweep_id\": \"${SWEEP_ID}\",
\"chat_session_id\": \"$CHAT_SESSION_ID\",
\"auto_start\": true
}"
Repeat once per grid combination.
curl -X GET "$SERVER_URL/runs" -H "X-Auth-Token: $AUTH_TOKEN"
curl -X GET "$SERVER_URL/sweeps/$SWEEP_ID" -H "X-Auth-Token: $AUTH_TOKEN"
curl -X GET "$SERVER_URL/wild/v2/system-health" -H "X-Auth-Token: $AUTH_TOKEN"
If a task requires experiment execution, do not proceed until the task plan includes explicit sweep/run API calls.
If you discover an existing run command that executes locally without API registration, convert it into a run creation call instead.