acceptodds
Under review as a conference paper at ICLR 2027

Automatically Evolving Prompt Guidelines for Task-Specific Optimization

Abstract

For Large Language Models to reliably answer user queries, users must clearly specify requirements, context, and constraints. In practice, however, user queries are often underspecified, forcing models to infer unstated assumptions that may misalign with the actual user intent. Existing prompt engineering guidelines aim to mitigate this issue, they are typically generic and task-agnostic, limiting their practical utility. Additionally, existing guidelines are formed manually and in a non-systematic way. To this end, we study prompt guideline optimization: the problem of automatically generating task-specific guidelines that help write better- specified prompts for a given task and model. Our key observation is that existing (completed) task examples (aka reference answers) often implicitly encode the missing information required to complete underspecified queries, including behav- ioral constraints, contextual assumptions, and evaluation criteria. We therefore propose AGOPS, an automatic approach that evolves task-specific guidelines via an optimization scheme that involves a prompt LLM writer, a solver LLM and prompt evolution, which maximize downstream effectiveness on a set of examples (user queries with reference answers). At inference time, our guidelines help to write well-specified prompts, boosting the effectiveness of LLMs. We show across mathematical reasoning, medical question answering, and coding tasks, that prompt underspecification leads to major drops (up to 95.3pp) in downstream task performance (compared to well-specified prompts). Following AGOPS guidelines can regain this loss (increasing performance up to 81.7pp) across all benchmarks.

Primary area:
generative models
Keywords:
prompt optimization, underspecification, abstention

Reviews

Among:
Shape:
1234this paper, Top 32%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 2.50 over 2 reviews, above 68% of the others.
Reviewer_5c6zOpenReview ↗
Rating
3(+1)
Confidence
3
Informativeness
2
Presentation
3
Soundness
2
Read review
Paper summary

This paper studies the problem of underspecified user prompts and proposes AGOPS, a framework for automatically learning task-specific prompt guidelines. Given an underspecified query, a reference answer, and a candidate guideline, a prompt-writer language model generates a more complete prompt. A solver model then answers the generated prompt, and its performance is used to evolve the guideline through GEPA. The framework uses character n-gram overlap constraints and rejection sampling to limit answer leakage. Experiments on mathematical reasoning, medical question answering, and coding benchmarks show substantial improvements over the selected baselines.

Critical strengths
  1. The method is clear and well structured. AGOPS automatically learns task-specific guidelines that help users identify what information should be included in their prompts. It uses reference answers to simulate more complete prompts and optimizes the guidelines through solver feedback and GEPA. The framework is relatively simple, and the learned guidelines can support both prompt writing and interactive clarification.

  2. The experimental results are strong across multiple benchmarks and models. AGOPS is evaluated on mathematical reasoning, medical question answering, and coding tasks using GPT-4.1-mini and Qwen3-32B. In the simulated setting, it consistently outperforms the selected baselines, with improvements of up to 81.7 percentage points over No Guideline. The experiments also show that the learned guidelines improve information seeking in multi-turn interactions (Figure 8). The additional model-scale experiments and user study provide further supporting evidence.

Critical weaknesses
  1. The main prompt-engineering evaluation gives the prompt writer access to reference answers for test instances (Appendix B.3), making it unclear whether the improvements would hold in real-world settings without such information. Furthermore, the assumption that reference answers contain the missing knowledge of users (Section 3.2) is not directly tested. The leakage constraint cannot reliably detect rewritten solutions, and Appendix F.2 explicitly instructs the writer to describe the exact algorithm in the reference answer. The No Guideline comparison also raises concerns. Figure 9 allows the writer to use reference answers only when explicitly instructed, but the initial guideline (Figure 10) does not include such instructions, while the optimized guidelines directly ask the writer to extract information from the reference answers. This makes it difficult to determine how much improvement comes from learning better guidelines rather than simply using more information from the reference answers. The small user study (20 participants, only 8 in the AGOPS group) also provides limited evidence of real-world effectiveness. More realistic evaluations and better-controlled baselines are needed.

  2. AGOPS combines existing ideas from instruction induction, guideline learning, and evolutionary prompt optimization. Related work, including "Guidelines to Prompt Large Language Models for Code Generation: An Empirical Characterization" (arXiv:2601.13118) and "Large Language Models Are Human-Level Prompt Engineers" (arXiv:2211.01910), explores similar ideas. However, the paper does not compare AGOPS with simpler methods that directly generate guidelines from reference examples without iterative optimization. Furthermore, the guidelines are optimized for specific tasks, but no experiments examine whether they can be used in different domains or adapted with only a few examples. It therefore remains unclear how well AGOPS works on new domains.

Questions for authors
  1. Some benchmarks remove nearly all information needed to answer the question, making refusing to answer a reasonable response rather than a failure. Although the paper reports abstention rates separately, the large accuracy drops do not distinguish between incorrect answers and reasonable refusals. An evaluation that rewards appropriate refusals or clarifying questions, together with more naturally underspecified prompts, would provide stronger evidence.

  2. Table 6 reports No Guideline performance at K=8 on the validation set (Appendix D.1), including 74.0 on GSM8K-Abstain, while Table 1 reports 8.9 for GPT-4.1-mini and 15.2 for Qwen3-32B on the test set. The Table 6 results are higher than both Table 1 results across all four datasets. The authors should clarify which solver, metric, and evaluation settings were used in Table 6 and explain these large differences. Furthermore, Table 7 reports an average improvement of +19.22, but the displayed values (91.36 and 72.41) give a difference of 18.95 percentage points. The use of reference answers and reasoning traces in Sections 3.1 and 3.2 could also be explained more clearly.

I would be willing to change my rating according to the authors' response.

Reviewer_mx2SOpenReview ↗
Rating
2
Confidence
2
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary
  • This paper proposes AGOPS, a framework for learning task-specific guidelines that help users supply missing information in their prompts. It uses a prompt writer, a solver, and GEPA to optimize guidelines based on downstream performance.
  • The prompt writer uses reference answers as a proxy for knowledge the user might possess. The framework includes filters intended to prevent the writer from directly exposing the answer.
  • Experiments cover underspecified math, medical QA, and coding tasks. The paper reports large gains in reference-assisted simulations, improvements in simulated clarification conversations, and a small user study suggesting that guidelines can improve human-written prompts.
Critical strengths
  • The studied problem is well-motivated and more specific than generic prompt polishing. The paper focuses on requirements that a user knows but may leave out, rather than assuming a more elaborate prompt can recover unavailable facts. Using GEPA to learn reusable specification guidelines is a sensible application of existing prompt-optimization methods.
  • The paper makes good effort to check the simulation. It includes a reference-assisted writer without optimized guidelines, leakage constraints and ablations, clarification experiments, and a user study where participants do not see reference answers.
Critical weaknesses
  • The largest gains are still an upper-bound result, not a demonstration of ordinary prompting performance. In the main simulation, the writer sees each test reference answer and uses it to fill in missing information. The No Guideline control shares that access, which is appropriate, but does not establish that real users can supply the same details. The overlap and reconstruction checks also do not rule out solution hints expressed in different words. I would be cautious about presenting the 81.7-point gain as the expected practical benefit.
  • The user study is too small to establish the added benefit of AGOPS over a baseline guideline. Appendix C analyzes twenty participants, with only eight in the AGOPS condition. Pass@1 improves from 51.3% to 75.0% for AGOPS and from 51.7% to 69.2% for the baseline, but these are different participant-authored tasks. AGOPS prompts are also longer. Without uncertainty estimates and a better-matched comparison, the study supports "guidelines help" more clearly than "these optimized guidelines are better."
Questions for authors
  • How much of the gain remains when a simulated user has access to the original complete problem specification but not the worked solution or reference implementation? The current clarification simulator also knows the expected answer, so this would be a useful less-privileged check.
  • Can you report participant-level uncertainty for the AGOPS-versus-baseline comparison, and clarify guideline assignment? A matched-task comparison would help separate guideline quality from task differences and extra prompt detail.
  • How many generated prompts were manually audited for adding specification versus revealing solution steps? Does the main gain remain on the subset judged not to contain solution-specific guidance?

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.