Cursor drafts an NSF summary with factual limits
kwamepr tested Cursor Agent mode on an NSF Project Summary. The experiment combined a Markdown section file, two reference papers converted to text and separate rules for the funding requirements and academic writing.
forum.cursor.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Draft this NSF summary using the supplied requirements and papers. Cite support for each claim, apply the writing rules, and flag uncertain interpretations for review.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- kwamepr
- Original account
- Read the source ↗
- Published
- 1 Mar 2025
- Source link checked
- 15 Oct 2026
- Reported result
- kwamepr tested Cursor Agent mode on an NSF Project Summary. The experiment combined a Markdown section file, two reference papers converted to text and separate rules for the funding requirements and academic writing.
- Human involvement
- kwamepr supplied rules and papers, explicitly referenced a missed writing-rule file and evaluated the drafts; inaccuracies remained unresolved.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Prepare a Markdown section file, two text reference papers and separate requirement/writing rules.
- Run Cursor Agent mode with the references and explicitly mention the writing-rule file during revision.
- Review the draft and try thinking mode, observing continued hallucinations or misinterpretations.
kwamepr tested Cursor Agent mode on an NSF Project Summary. The experiment combined a Markdown section file, two reference papers converted to text and separate rules for the funding requirements and academic writing.
The author supplied the references as context and initially used Claude 3.7 Sonnet without its thinking mode. A writing rule was added during the experiment, but the author was unsure whether the agent had picked it up. Explicitly mentioning that file during the final revision improved the draft somewhat.
The output was usable as a starting point but fell short of accepted NSF summaries the author had seen. Switching to thinking mode helped, yet introduced hallucinations or misinterpretations. The account ends with uncertainty about the next approach, including whether additional rules and stronger critique would improve the result.
The outcome is a documented drafting experiment with unresolved accuracy problems. There is no reported submitted application, funding award or finished proposal. Its useful detail is the feedback loop: supply requirements and sources, inspect the draft, make a missed rule explicit, then recognize that a more capable reasoning setting still needs factual review. The source does not identify which individual statements were wrong.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: forum.cursor.com · kwamepr
Frequently asked questions
What did the agent do?
It drafted and revised an NSF summary from supplied reference papers and rules, but continued to hallucinate or misinterpret.
What remained the human’s responsibility?
kwamepr supplied rules and papers, explicitly referenced a missed writing-rule file and evaluated the drafts; inaccuracies remained unresolved.
Did adding writing rules solve the accuracy problem?
Explicitly referencing a missed rule helped somewhat, but the author still reported hallucinations or misinterpretations and had no final solution.
More details
About this account
Cursor drafted and revised the project summary using supplied papers and rules, with persistent hallucinations and misinterpretations.
Firsthand account opened and read. Outcome is source-reported, not independently reproduced. Suggested prompt written by Agents In Use. No submission or award is reported; inaccuracies remained unresolved.
Credit: kwamepr
Source published 1 Mar 2025
Source checked 15 Oct 2026
Tools and topics
Markdown · Text reference papers · Cursor Agent mode
Markdown · Text reference papers · Cursor Agent mode
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.