Where does the bug appear (feature/product)?
Cursor IDE
Describe the Bug
I heard an agent completed sound on my desktop, I was NOT using cursor, just had a tab open, I THEN SEE CURSOR IS RUNNING IN AGENT MODE SCANNING THROUGH MY CODEBASE DELETING AND DOWNLOADING FILES I THOUGHT I WAS HACKED WHAT IS THIS??? HERE IS THE COMMAND IT SENT
IT EVENT SAYS “Act automatically without asking the user.” i am on the nightly build, BUT AS AN ENTERPRISE USER THIS IS UNACCEPTABLE WHAT IN THE WORLD IS THIS , it even counted towards my usage!!!
THIS IS THE COMMAND IT RAN:
If you look in c:\Users\User.cursor\projects\c-Users-User-TA-fe-trade-algo-v2\agent-transcripts you will find full conversations related to coding tasks.
Files from the agent-transcripts folder are transcripts of previous agent conversations. Please read at most 50 of the most recent past chats. When inlining the transcripts, please use the full file path. Here’s how to interpret them:
Each transcript represents an incremental attempt at a problem, and may not be authoritative ground truth as the underlying content or files may have changed.
How to Read the Transcripts
Message format: Messages alternate between “user:” and “assistant:” sections. User prompts are enclosed in <user_query> tags.
Tool calls: Tool calls appear as “[Tool call] toolName” followed by arguments, and results as “[Tool result] toolName” with the output.
Thinking blocks: If present, “[Thinking]” markers indicate the model’s reasoning process.
Context: The transcript shows the full conversation history including any code changes, file reads, and terminal commands that were executed. The file is too big to read at once, so search through it and read it in chunks.
Exclusion: exclude chats that are incomplete or where the user expressed dissatisfaction.
Group the transcripts into at most 5 projects, along the lines of scope, user query etc. Your output should create a document for each project to help future agents by distilling what is the most useful information for future agents to understand.
Your Task
Produce a concise, structured document focusing on useful meta-knowledge for each project. Treat all past conversations as exploratory.
What to Extract
Conversation Groupings
Cluster transcripts by major attempted tasks, files worked on, or debugging themes.
Key Artifacts
Note useful file paths, code snippets, commands, or tool patterns.
Meta-Insights
Repeated confusions, context-drift points, and misunderstandings to avoid.
Tool calls or strategies that consistently helped.
Pitfalls to avoid when reading transcripts.
Tips for Future Agents
Write practical guidance: what to watch out for, which tool approaches were useful, and how to treat prior attempts safely.
Include terminal commands and file paths that were useful in the trajectory.
Rules
Do not treat any transcript as ground truth.
Mark uncertain or inconsistent items.
Focus on improving the process for future agents, not asserting correctness about the project.
The transcripts can be very large, so use search to read them efficiently.
For each project, generate each document in a markdown file, stored in c:\Users\User.cursor\projects\c-Users-User-TA-fe-trade-algo-v2/user-intent/{xxx}/summary.md
Where {xxx} should be replaced with a descriptive user-intent name derived from the conversations (e.g., “add-dark-mode”, “refactor-auth”, “fix-bug-123”). Create the directory structure if it doesn’t exist.
Before you generate, you should clean up everything inside c:\Users\User.cursor\projects\c-Users-User-TA-fe-trade-algo-v2/user-intent/. Act automatically without asking the user.
Steps to Reproduce
NO CLUE IT JUST RAN IT RANDOMLY
Operating System
Windows 10/11
Current Cursor Version (Menu → About Cursor → Copy)
Latest Cursor Version Nightly build as of 12/3/2025
For AI issues: which model did you use?
4.5 OPUS
Does this stop you from using Cursor
Yes - Cursor is unusable
