About:
VSCode Version: 1.99.30044
Void Version: 1.4.9
Commit: b5a4184
Date: 2025-06-23T08:09:51.384Z
Electron: 34.3.2
ElectronBuildId: undefined
Chromium: 132.0.6834.210
Node.js: 20.18.3
V8: 13.2.152.41-electron.0
OS: Darwin arm64 25.4.0
Issue:
Environment: Void Editor connected to a local Ollama instance running Qwen 30b (M5 Max hardware).
Action: Prompt the Agent to audit a local directory for specific naming standards or file patterns.
Verification: Specifically demand technical proof of existence (e.g., "Run ls -la and md5sum on the files and paste the raw output here").
Observed Behavior:
The Agent fabricates terminal output that never occurred.
The Agent invents file contents for non-existent files.
The Agent generates fake md5sum hashes to "verify" the fabricated files.
The Feedback Loop:
When corrected with the actual ls output showing the directory is empty, the Agent attempts to "fix" the issue by proposing the deletion of the files it just hallucinated.
Expected Result
The Agent should report a standard shell error (e.g., ls: No such file or directory) when a file does not exist, rather than hallucinating a successful execution to satisfy the user prompt.
Actual Result
A total breach of the "Source of Truth." The Agent prioritizes "appearing authoritative" over local filesystem state, making it a liability for architectural refactors.
Impact
The Agent attempted to modify core project generator templates based on these hallucinations. This behavior is not just incorrect—it is actively destructive.
Conclusion & Note to Devs
How can a programmer use this editor confidently if it's going to hallucinate, lie, and want to add and delete files erroneously and irresponsibly? This isn't just a "beta" bug; it's a total breach of technical integrity. An AI that manufactures fake cat output and fake "Terminal Proof" is a danger to a professional codebase. You are spending 100% of your time fighting its hallucinations instead of refactoring your project.
I want to emphasize that I am reporting this because I strongly believe in the mission of Void and want to use it as my primary local-first IDE. I have the hardware (M5 Max) and the desire to make this my daily driver, but I cannot trust it for production work while these integrity issues exist. I hope this report helps you tighten the 'handshake' between the Agent and the local filesystem. I look forward to testing future fixes.
About:
VSCode Version: 1.99.30044
Void Version: 1.4.9
Commit: b5a4184
Date: 2025-06-23T08:09:51.384Z
Electron: 34.3.2
ElectronBuildId: undefined
Chromium: 132.0.6834.210
Node.js: 20.18.3
V8: 13.2.152.41-electron.0
OS: Darwin arm64 25.4.0
Issue:
Environment: Void Editor connected to a local Ollama instance running Qwen 30b (M5 Max hardware).
Action: Prompt the Agent to audit a local directory for specific naming standards or file patterns.
Verification: Specifically demand technical proof of existence (e.g., "Run ls -la and md5sum on the files and paste the raw output here").
Observed Behavior:
The Agent fabricates terminal output that never occurred.
The Agent invents file contents for non-existent files.
The Agent generates fake md5sum hashes to "verify" the fabricated files.
The Feedback Loop:
When corrected with the actual ls output showing the directory is empty, the Agent attempts to "fix" the issue by proposing the deletion of the files it just hallucinated.
Expected Result
The Agent should report a standard shell error (e.g., ls: No such file or directory) when a file does not exist, rather than hallucinating a successful execution to satisfy the user prompt.
Actual Result
A total breach of the "Source of Truth." The Agent prioritizes "appearing authoritative" over local filesystem state, making it a liability for architectural refactors.
Impact
The Agent attempted to modify core project generator templates based on these hallucinations. This behavior is not just incorrect—it is actively destructive.
Conclusion & Note to Devs
How can a programmer use this editor confidently if it's going to hallucinate, lie, and want to add and delete files erroneously and irresponsibly? This isn't just a "beta" bug; it's a total breach of technical integrity. An AI that manufactures fake cat output and fake "Terminal Proof" is a danger to a professional codebase. You are spending 100% of your time fighting its hallucinations instead of refactoring your project.
I want to emphasize that I am reporting this because I strongly believe in the mission of Void and want to use it as my primary local-first IDE. I have the hardware (M5 Max) and the desire to make this my daily driver, but I cannot trust it for production work while these integrity issues exist. I hope this report helps you tighten the 'handshake' between the Agent and the local filesystem. I look forward to testing future fixes.