inference-scaling

name: inference-scaling description: "Use when the user wants to run inference-time scaling on a prompt — detect environment, execute scaling, and present results. For algorithm selection, budget tuning, reward models, and troubleshooting, consult the inference-scaling-guide skill." allowed-tools: ["Bash(${CLAUDE_PLUGIN_ROOT}/scripts/its_scale.sh:)", "Bash(${CLAUDE_PLUGIN_ROOT}/scripts/its_detect.sh:)"]

Run Inference-Time Scaling

Execute scaling on a prompt. For algorithm guidance, budget tuning, and troubleshooting, consult the inference-scaling-guide skill.

Step 1: Check Environment

"${CLAUDE_PLUGIN_ROOT}/scripts/its_detect.sh"

If not ready

• library=missing or config=missing: invoke the setup-guide skill.

If ready (library=installed, config=found)

Proceed to Step 2.

Step 2: Execute Scaling

Run the scaling script with the user's prompt and any overrides:

"${CLAUDE_PLUGIN_ROOT}/scripts/its_scale.sh" --metadata $ARGUMENTS

If the user provides a file path (e.g., "scale all prompts in data/eval.jsonl"), invoke the batch-scaling skill instead.

Step 3: Present Results

1. Selected response — Show the winning response prominently
2. Metadata (if available):
   • Self-consistency: show vote counts ("Selected by majority vote — 5/8 responses agreed")
   • Best-of-N: show scores ("Selected as highest scoring — score: 0.92 out of 8 candidates")
3. Configuration used — algorithm, budget, model (briefly)

If the scaling failed, consult the inference-scaling-guide skill for troubleshooting.