Overview

1. One thing at a time. Each response advances exactly one decision or collects one piece of information.
2. Confirm before proceeding. Wait for the user to approve the spec before considering this skill complete.
3. Infer, don’t interrogate. Use what’s already known from the conversation. Only ask when you truly can’t infer.
4. Do NOT ask about base model selection. Model selection is handled exclusively by the model-selection reference.

Workflow

Step 0: Check for Existing Spec

Before starting discovery, check if a *_use_case_spec.md file already exists in the project. If it does, present it to the user and ask whether they want to reuse it, modify it, or start fresh.

Step 1: Determine Intent

Check the plan (PLAN.md) or conversation context to determine whether the user wants to:

• Fine-tune a model → read references/spec-for-finetuning.md and follow it.
• Deploy a base model → read references/spec-for-deployment.md and follow it.

If the intent is already clear from the plan (e.g., the plan includes finetuning steps vs. only model-selection + model-deployment), use that. If ambiguous and not already resolved by the planning skill, ask:

“Are you looking to fine-tune a model for your use case, or deploy a base model as-is?”

⏸ Wait for user response.

Edit Protocol

• If the user requests changes pertaining to any information covered by use_case_spec.md, you must edit it accordingly and ask for confirmation again.
• The user can edit use_case_spec.md directly if they want to. If the user says they’ve updated the file directly, read it to get the latest in your context.

References

• references/spec-for-finetuning.md — Discovery and spec generation workflow for fine-tuning
• references/spec-for-deployment.md — Discovery and spec generation workflow for base model deployment