Manage files in a chosen folder with an agent
The author built an open-source desktop agent that works inside a chosen folder, reading and writing files and asking before changing anything. Task: Suggested implementation: Build a desktop agent restricted to a user-selected folder. The agent should read and write text files, requiring explicit user approval for every modification. Shell command execution must be disabled by default, enabled only via manual user override. Human role involves selecting the target folder and approving each file change.; tools: Claude Code. Human role: The original post does not say how a person stayed involved.; limitations: Prerequisites include a local environment supporting the chosen LLM provider (Claude, OpenAI, Google, or local). Constraints: Sandboxing may be incomplete, and the agent is limited to text files. Operating costs depend on API usage or local hardware. Integration uncertainty exists regarding specific file system permissions. Human approval is mandatory for all write operations..
Tools
Published on Agent Work Atlas:
Overview
The author built an open-source desktop agent that works inside a chosen folder, reading and writing files and asking before changing anything.
Workflow
Suggested implementation: Build a desktop agent restricted to a user-selected folder. The agent should read and write text files, requiring explicit user approval for every modification. Shell command execution must be disabled by default, enabled only via manual user override. Human role involves selecting the target folder and approving each file change.
Reported result
Actual outcomes are unknown as the source is unavailable. Proposed acceptance check: Verify that the agent successfully reads and writes text files within the selected folder only. Confirm that no file changes occur without explicit user approval and that shell commands remain disabled unless manually enabled.
Human role
The original post does not say how a person stayed involved.
Limitations
Prerequisites include a local environment supporting the chosen LLM provider (Claude, OpenAI, Google, or local). Constraints: Sandboxing may be incomplete, and the agent is limited to text files. Operating costs depend on API usage or local hardware. Integration uncertainty exists regarding specific file system permissions. Human approval is mandatory for all write operations.
Try it
I want to test a folder-restricted desktop agent. [replaceable inputs: target folder path, LLM provider (e.g., Claude/OpenAI/local), sample text files]. Steps: 1. Install the agent and configure it to access only [target folder path]. 2. Enable read-only mode and list files to verify access. 3. Attempt to create a new text file named 'test.txt' in the folder. 4. Confirm the agent requests approval before writing. 5. Approve the change and verify the file exists. 6. Attempt to execute a shell command and confirm it is blocked by default. Quality metric: Ensure no files are modified outside the target folder and all write operations require explicit user confirmation. If inputs are missing, ask for the folder path and LLM provider. Do not proceed with real file modifications until I confirm the read-only test passed.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.