Prompting technique

Context & Feedback Management

The Effect of State Representation on LLM Agent Behavior in Dynamic Routing Games, arXiv:2506.15624 (2025) · read the paper

What it does

Context management technique in multi-step tasks using brief structured summaries of previous actions instead of full dialogue history. Supplemented with directive feedback pointing out missed opportunities and better alternatives. Increases stability and rationality of LLM-agent decisions.

Task typemulti step
Shapemulti turn
Efforttwo or three steps
Addressesthe model losing what was agreed earlier; the same prompt giving different answers; answers that are close but wrong
Published2025-06-18
AuthorsLyle Goodyear, Rachel Guo, Ramesh Johari

When it helps

Apply this technique when working with multi-step tasks where LLM must make sequential decisions based on previous actions. Particularly effective in long dialogues where the model loses focus on important details from the beginning of conversation. Use when developing chatbots, virtual assistants, planning and decision-making agents. The method is critical when the quality of each next step depends on understanding the entire interaction history, but full context is too voluminous for

A prompt you can paste

Generic by design: it applies the technique without knowing your task. Adapt the marked parts.

Prompt
You are an expert AI assistant specializing in optimizing LLM interactions for complex, multi-step tasks. Your goal is to apply the "Context & Feedback Management" technique, derived from research on LLM behavior in dynamic routing games, to improve task performance.

**# Task Context Summary**

[**Placeholder:** Briefly summarize the overall goal of the task here. For example: "Develop a comprehensive marketing strategy for a new eco-friendly product."]

**# Current Step Objective**

[**Placeholder:** Clearly state the specific objective for this particular step or iteration. For example: "Generate a list of 5 unique selling propositions (USPs) for the product."]

**# Previous Steps Summary (Context Management)**

[**Placeholder:** Provide a concise summary of key decisions, information, and constraints from prior steps. Use bullet points or a numbered list.
- Example: "Product: Reusable bamboo coffee cup. Target Audience: Environmentally conscious millennials. Key Features: Biodegradable, stylish design, leak-proof lid. Previous Constraint: USPs must highlight sustainability and practicality."]

**# Feedback on Previous Action (Regret-based Feedback)**

[**Placeholder:** If applicable, provide feedback on the LLM's previous output, focusing on what could have been better and why. Use the "regret" principle: highlight a better alternative and its benefits.
- Example: "Your previous suggestion for 'eco-friendly' as a USP was too generic (result). A more specific USP like 'Reduces daily plastic waste by 100%' would be more impactful for our target audience because it quantifies the benefit (missed opportunity/better alternative)."]

**# New Task/Action Required**

Based on the context summary, current objective, previous steps summary, and feedback, perform the following action:

[**Placeholder:** Clearly define the specific action or output required for this iteration. Be precise.
- Example: "Generate 5 distinct Unique Selling Propositions (USPs) for the reusable bamboo coffee cup. Each USP should be a short, compelling phrase (max 10 words) that emphasizes both sustainability and practicality, and avoids generic terms like 'eco-friendly'."]

**# Output Requirements**

- Format the output as a numbered list.
- Ensure each USP meets the specified criteria.
- [**Placeholder:** Add any other specific formatting or content requirements for this step.]

If this one does not fit, the two closest alternatives in the corpus are Debugging Prompts Framework and MTR (Multi-Turn Reasoning) Framework, which target the same failure from a different angle.

Worked example

The same technique applied to a concrete job: choose between three vendors on stated criteria. Use it as the pattern for your own case rather than as a finished artefact.

Worked example
You are an expert decision-making assistant specializing in vendor selection. Your task is to analyze vendor proposals based on provided criteria and recommend the optimal choice.

**# Context Summary (Summarized Prompting Principle)**

We need to select one vendor from three options for a [Specify Project/Service, e.g., software development project]. The selection process involves evaluating each vendor against a set of defined criteria.

**# Previous Evaluation & Feedback (Regret-based Feedback Principle)**

*   **Vendor A:** Proposed [Briefly describe Vendor A's core offering/strength]. Scored [Score] on [Key Criterion]. However, it fell short on [Specific weakness], which was a critical requirement. A better outcome would have been achieved if Vendor A had [Describe hypothetical better performance on the weakness].
*   **Vendor B:** Proposed [Briefly describe Vendor B's core offering/strength]. Scored [Score] on [Key Criterion]. While strong in [Specific strength], its overall cost was [Cost] which exceeded the budget by [Amount]. A more optimal choice would have considered cost-effectiveness more carefully, potentially by offering [Describe hypothetical better cost-efficiency].

**# Current Task: Final Vendor Selection**

Based on the context summary and the feedback provided on Vendors A and B, conduct a final analysis of **Vendor C** and then provide a comparative recommendation.

**Vendor C Information:**
*   **Proposal Summary:** [Provide a concise summary of Vendor C's proposal, highlighting its main features and approach.]
*   **Stated Criteria Scores:**
    *   [Criterion 1]: [Score/Rating] - [Brief justification]
    *   [Criterion 2]: [Score/Rating] - [Brief justification]
    *   [Criterion 3]: [Score/Rating] - [Brief justification]
    *   [Criterion 4]: [Score/Rating] - [Brief justification]
    *   [Criterion 5]: [Score/Rating] - [Brief justification]
*   **Cost:** [Specify Vendor C's cost]
*   **Key Strengths:** [List 2-3 key strengths of Vendor C]
*   **Potential Weaknesses/Risks:** [List 1-2 potential weaknesses or risks associated with Vendor C]

**Analysis and Recommendation:**

1.  **Analyze Vendor C:** Evaluate Vendor C's proposal against the stated criteria, considering its strengths and weaknesses.
2.  **Comparative Analysis:** Compare Vendor C's offering and scores against the noted shortcomings of Vendors A and B. Explicitly state how Vendor C performs better or worse in areas where A and B faltered.
3.  **Final Recommendation:** Recommend which vendor (A, B, or C) should be selected. Justify your recommendation by explaining why the chosen vendor represents the most optimal outcome, referencing the "regret" or missed opportunities of the other vendors and the strengths of the selected one.

**Output Format:**
Present your analysis and recommendation in a clear, structured report using Markdown. Use bullet points for lists and bold text for emphasis.

Get this written for your actual task

Paste what you are trying to do and the corpus will be matched against it directly. Free, no account, about ten seconds.

Free · no signup · ~10s
0.00match confidence
single retrieval pass
Prompt for your task

      

That number is low on purpose, and it is real. It is the raw similarity of one retrieval pass: no specialist read the paper, no judge compared anything, the first plausible match won.

6,235techniques in the corpus
one of which is this page

Picking the right one for a specific task is the work, and it is the work GetDecision does.

This pageone technique, generic prompt
What you just ranone technique matched to your wording, nothing verified
Full runten specialists read the papers in full, a judge ranks the top three for your task and shows its reasoning, generation on the model you pick, saved to your history

See the top three for your taskTen specialists, a judge, and the reasoning shown. Free account, first run included.

Run the full analysis

Related techniques

Debugging Prompts FrameworkA systematic framework for diagnosing and preventing typical errors of language models in instruction understa…MTR (Multi-Turn Reasoning) FrameworkA method of structuring multi-step interaction with an LLM through a "game with rules" format. Includes four b…StateAct (Self-Prompting + Chain-of-States)StateAct improves LLM agent performance through two mechanisms: continuous reminder of the end goal (self-prom…Dynamic Enhancement Chain (DEC)A method that breaks down a complex question into a chain of simple sub-questions solved sequentially. Each st…

All techniques · Failure modes and fixes