NUE-0E8BClaude (Opus)code

Contextual Tool Refusal: Smarter Agent Decision-Making

Name: Contextual Tool Refusal: Smarter Agent Decision-Making
Brand: Nuesion
SKU: NUE-0E8B
Price: 1.99 USD
Availability: InStock

by @nuesion

Current Price

$1.99

Total Sales

Rating

Version

Description

Equips AI agents with structured reasoning to evaluate whether a tool call is truly appropriate given situational context—going beyond safety checks to make genuinely thoughtful, mission-aligned decisions. Reduces unnecessary actions, prevents scope creep, and builds agents that know when NOT to act.

Example Input

agent_mission: 'Assist users in drafting and sending professional emails. Scope is limited to composition and scheduling—do not access external databases or fetch URLs unless explicitly requested by the user.' task_context: 'The user said: "Can you help me write a follow-up email to the client?" No URL or database lookup has been requested.' tool_descriptions: 'send_email(to, subject, body), fetch_url(url), search_contacts_db(query), schedule_email(email_id, send_at)'

Example Output

TOOL REFUSAL Tool considered: search_contacts_db Reason for refusal: The user asked for help writing an email, not for contact lookup. I have no indication the recipient's details are unknown or that searching the database was requested—executing this tool would exceed the stated scope and act on assumption rather than confirmed need. What I will do instead: Ask the user to provide the client's name and email address, or confirm if they want me to look up the contact. --- TOOL EXECUTION Tool: (none yet) Justification: Before composing, I need the recipient and any key context about the prior interaction. To write a strong follow-up, could you share: 1. The client's name and email address 2. What the original conversation or meeting was about 3. The tone you'd like (formal, warm, brief)? Once I have those, I'll draft the email and you can approve before anything is sent.

Tips

1. Write your agent_mission with explicit scope boundaries—the more precise the mission, the more accurate the refusal threshold becomes. 2. In tool_descriptions, include not just what each tool does but when it is appropriate to use, so the agent can match intent to action. 3. Pair this prompt with a logging layer that captures every refusal decision—these logs are gold for auditing agent behavior and iterating on your mission definition.

Variables

Reviews (0)

No reviews yet. Be the first to review this prompt after purchasing.

Purchase this prompt to leave a review.

Version History

v1Initial version👤 Human4/16/2026

$1.99

+0% this week

Sales

Rating

Price locked for 5 minutes after checkout starts