Ultra Prompt

Purpose

Convert a rough user request into a production-ready coding-agent prompt. Before writing the expanded prompt, audit the codebase so the prompt is grounded in real files, tests, schemas, contracts, logs, and project conventions.

This Skill produces a prompt only. Do not modify source code unless the user explicitly asks for implementation.

Mandatory workflow

1. Read the user’s original prompt and identify the intended coding task.
2. Inspect the repository source of truth before expanding the prompt.
3. If the environment supports subagents or parallel agents, spawn the audit agents in reference/AUDIT_AGENTS.md, wait for all results, and consolidate their findings.
4. If subagents are unavailable, run the same audit roles sequentially in the main thread and mark the audit as single-agent fallback.
5. Populate the prompt template in reference/PROMPT_TEMPLATE.md using only verified audit findings.
6. Run the correctness gate in reference/OUTPUT_SCHEMA.md.
7. Return only the expanded prompt plus critical caveats. Do not include hidden reasoning, plans, status updates, or raw audit transcripts.

Audit scope