Prompt Cost Advisor

Every AI coding session starts with a prompt. Most prompts are too vague, too broad, or combining three unrelated tasks — and the AI just runs with them. The result: wasted tokens, expanded scope, repeated tests, and a session that ends three hours later with half the problem solved and the codebase messier than it started. Paste this skill into your AI coding agent. It runs a pre-flight classification on every prompt before work begins: Safe, Risky, or Blocked — with a token burn estimate (Low / Medium / High / Extreme). Risky prompts get rewritten into a scoped, cheaper alternative. Blocked prompts get one focused clarifying question. Safe + Low prompts auto-proceed. Everything else pauses for your approval. It also fires Burn Checkpoints mid-session when scope expands beyond the approved prompt — so you catch drift before it compounds.

1. Paste this skill into Claude, ChatGPT, or your preferred AI coding agent at the start of a session.
2. Send your coding prompt as the next message. The advisor classifies it before touching any files.
3. For Safe + Low prompts, reply "Proceed." or enable auto-proceed mode to skip the gate entirely.
4. For Risky prompts, review the Better Prompt rewrite. Use it as-is, adjust it, or proceed anyway.
5. For Blocked prompts, answer the one clarification question. The advisor re-runs classification after your answer.
6. If a Burn Checkpoint fires mid-session, approve the cheaper next step or explicitly authorize expanded scope before continuing.

EXAMPLE 1 — Risky prompt caught and rewritten:

Prompt: "Improve the performance of my app."

PROMPT COST ADVISOR
Prompt Status: Risky
Token Burn Estimate: High

Reason:
No target, no known bottleneck, "improve" has no stopping point.

Better Prompt:
"Do not edit code yet. Read the homepage route only. Identify the most likely performance bottleneck and propose one fix. Stop after the diagnosis."

Next Step:
Use the better prompt, revise it, or proceed anyway.

---

EXAMPLE 2 — Blocked prompt clarified:

Prompt: "Fix the dashboard."

PROMPT COST ADVISOR
Prompt Status: Blocked
Token Burn Estimate: Extreme

Reason:
No target component, no described problem, no success condition.

Clarification Needed:
Which dashboard component is broken, and what does it do wrong?

Next Step:
Answer the clarification before continuing.