Issue #11 - [FEATURE] Enhance ai command to execute shell commands from AI responses

[rogelio-cmj]

Motivation

The current ai command in the Puter shell can only respond with text-based answers to user queries. However, when users ask the AI to perform actions like "create a directory named test" or "list all files", the AI can only describe what command to run rather than actually executing it. This creates friction in the user experience, requiring users to manually copy and execute the suggested commands.

By enabling the ai command to recognize and execute shell commands directly, we can create a more seamless and powerful interface where users can interact with their system using natural language. This bridges the gap between conversational AI and system automation, making the shell more accessible to users who may not remember exact command syntax.

Current Behavior

The ai command currently only outputs text responses from the AI model. When a user asks the AI to perform an action that requires a shell command, the AI can only respond with a textual description or suggestion, but cannot actually execute the command.

Reproduction Steps:

1. Open the Puter shell terminal. (If you don't see the terminal app, go to the App center and search)
2. Run the command: ai "create a directory named test"
3. Observe: The AI responds with text explaining how to create a directory or suggesting the command, but does not actually create the directory
4. Run ls to verify that no directory was created
5. Expected: The directory should have been created automatically
6. Actual: Only a text response is provided, no directory exists

Expected Behavior

The ai command should be able to detect when the AI response contains a shell command that should be executed, extract that command, prompt the user for confirmation, and execute it if confirmed. The AI should be instructed via system prompts to wrap executable commands in a special marker format that can be parsed and extracted.

Acceptance Criteria:

• The AI receives system instructions to wrap executable commands in a special marker format (e.g., %%%command%%%)
• The ai command parses AI responses to detect and extract wrapped commands using pattern matching
• When a command is detected, the user is prompted for confirmation before execution (y/n)
• If the user confirms, the command is executed using the shell's pipeline execution mechanism
• If the user declines or provides any response other than 'y', the command execution is cancelled with an appropriate message
• Regular text responses (without commands) continue to work as before, displaying the AI's message
• The command input stream is configured to read user confirmation synchronously

Verification

Manual Testing:

1. Run ai create a directory named mytest in the Puter shell
2. Verify that the AI responds with a message and a command wrapped in markers
3. Verify that you are prompted: Execute command: 'mkdir mytest' (y/n):
4. Type y and press Enter
5. Verify that the command executes and the directory is created
6. Run ls to confirm the mytest directory exists
7. Run ai what is the weather like (a non-command query)
8. Verify that the AI responds with text only, no command execution prompt
9. Run ai list all files and type n when prompted
10. Verify that the command is cancelled and not executed

Testing Edge Cases:

1. Test with various command types (mkdir, ls, cat, etc.)
2. Test declining command execution to ensure cancellation works
3. Test queries that don't require commands to ensure normal behavior is preserved
4. Verify error handling when a command fails to execute

Submission

Download https://cap.so/ to record your screen (use Studio mode). Export as an mp4, and drag and drop into an issue comment below.

Guide to submitting pull requests: https://hackmd.io/@timothy1ee/Hky8kV3hlx