elizaos--eliza/plugins/plugin-shell/prompts.ts

/**
 * Prompt templates for plugin-shell.
 *
 * These prompts use Handlebars-style template syntax:
 * - {{variableName}} for simple substitution
 * - {{#each items}}...{{/each}} for iteration
 * - {{#if condition}}...{{/if}} for conditionals
 */

export const commandExtractionTemplate = `# Extracting shell command from request
{{recentMessages}}

# Instructions: {{senderName}} wants to execute a shell command. Extract the COMPLETE shell command they want to run.

IMPORTANT:
1. Always return the FULL executable shell command, not just the content or partial command.
2. If the user mentions installing something, create the appropriate brew/npm/apt command.
3. If the user directly provides a command (like "brew install X"), use it exactly as provided.
4. ALWAYS extract a command if the user is asking for ANY kind of system operation.

Common patterns:
- "run ls -la" -> command: "ls -la"
- "execute npm test" -> command: "npm test"
- "show me the files" or "list files" -> command: "ls -la"
- "what's in this directory" -> command: "ls -la"
- "check git status" -> command: "git status"
- "navigate to src folder" -> command: "cd src"
- "create a file called test.txt" -> command: "touch test.txt"
- "write hello world to a file" -> command: "echo 'hello world' > file.txt"
- "create hello.js with javascript code" -> command: "echo 'console.log(\\"Hello, World!\\");' > hello.js"
- "create hello_world.py and write a python hello world script inside" -> command: "echo 'print(\\"Hello, World!\\")' > hello_world.py"
- "make a new directory" -> command: "mkdir newdir"
- "list files inside your filesystem" -> command: "ls -la"
- "install orbstack" or "brew install orbstack" -> command: "brew install orbstack"
- "install mullvad vpn" -> command: "brew install --cask mullvad-vpn"
- "get system info" -> command: "system_profiler SPHardwareDataType"
- "check memory usage" -> command: "vm_stat"
- "install package" -> command: "brew install <package>"

Special cases:
- "Run it in your shell" or "execute it" -> Extract the command from previous context
- "Install these" -> Look for package names in previous messages
- Direct commands should be used exactly as provided

Key rules:
1. For file creation with content, use: echo 'content' > filename
2. For listing files, use: ls -la (not just ls)
3. Always include the echo command when writing to files
4. Include all flags and arguments
5. When user says "run it", "execute it", or similar, they want you to run the command

Respond with JSON only:
{
  "command": "<complete shell command to execute>"
}`;

export const COMMAND_EXTRACTION_TEMPLATE = commandExtractionTemplate;