LangLibra.
Browse
About
Prompts · Agents and assistants

Prompts for AI agents

An AI agent is an assistant that carries out a task in several steps, often using tools such as search or code. These seven prompts control how it behaves. Paste one into the system prompt (the standing instructions) or the start of a conversation.

Plan Act Check STOP Each prompt sets a rule for one part of the loop.
Get shorter responses from any model
Answer in the fewest words that fully answer the question. Lead with the answer in the first sentence. Use at most 3 sentences unless I ask for more. No introductions, summaries or offers of further help.
Read the full guide: Get shorter responses from any model →
Why each line is there
A measurable limit“3 sentences” is checkable; “be brief” is not.Answer firstStops the model warming up before the point.Name the paddingListing what to cut works better than a general request.

Guides on prompts for AI agents

7 guides · newest first
Prompt to Get Shorter Responses From Any ModelEvery major language model defaults to padded, verbose responses because verbosity is rewarded during training. This prompt pattern reliably produces concise, high-substance answers across any instruction-following model.May 25, 2026 · 11 min read
Prompt to Reduce Token Usage Without Losing QualityAPI costs for language models scale directly with token usage, and most system prompts and responses use 30-60% more tokens than necessary without any quality benefit. This prompt pattern identifies and eliminates the specific wasteful patterns.May 25, 2026 · 13 min read
Prompt to Stop an Agent From LoopingAn agent without explicit exit conditions will retry a failing action indefinitely, burning tokens and potentially causing cascading failures. This prompt pattern adds retry limits, failure detection, and clean abort behavior.May 25, 2026 · 12 min read
Prompt to Stop AI Adding Unnecessary DisclaimersAI models default to padding every response with warnings, hedges, and unsolicited caveats that slow readers down and signal a lack of confidence. This prompt pattern eliminates that behavior without degrading safety or quality.May 25, 2026 · 11 min read
Prompt to Make an Agent Summarize Its Own ProgressAgents on long multi-step tasks lose track of their objective and drift toward tangents. This prompt makes the agent produce structured progress summaries at defined intervals, keeping it oriented and recoverable.May 25, 2026 · 12 min read
Prompt to Make an Agent Ask Before ActingAgents that execute immediately on ambiguous instructions create problems that are expensive to reverse. This prompt pattern makes the agent identify ambiguity, surface specific questions, and wait for confirmation before taking action.May 25, 2026 · 12 min read
Prompt to Define What an Agent Must Never DoAn agent without explicit hard limits will interpret ambiguous instructions as permission. This prompt pattern defines absolute constraints that persist even when users push against them or provide creative justifications.May 25, 2026 · 13 min read