diff --git a/shellgenius/generation.py b/shellgenius/generation.py
new file mode 100644
index 0000000..6304c84
--- /dev/null
+++ b/shellgenius/generation.py
@@ -0,0 +1,347 @@
+"""Shell generation engine for ShellGenius."""
+
+import logging
+import re
+from dataclasses import dataclass
+from typing import Any, Dict, List, Optional, Tuple
+
+from shellgenius.config import get_config
+from shellgenius.ollama_client import get_ollama_client
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class GeneratedScript:
+    """Container for generated shell script."""
+
+    commands: List[str]
+    explanation: str
+    shell_type: str
+    raw_response: str
+
+
+class PromptTemplates:
+    """Prompt templates for shell generation."""
+
+    BASH_GENERATE = """You are a shell script expert. Generate a {shell} script for the following task:
+
+Task: {description}
+
+Requirements:
+- Use safe, best practices
+- Add comments explaining each command
+- Use proper error handling with set -e or similar
+- Return ONLY the script content, no markdown code blocks
+
+Script:"""
+
+    BASH_EXPLAIN = """You are a shell script expert. Explain this {shell} script line by line:
+
+{script}
+
+Provide explanations in this format:
+Line <num>: <command> - <explanation>
+
+Be concise and informative:"""
+
+    BASH_REFACTOR = """You are a shell script security expert. Analyze this {shell} script for security issues and suggest safer alternatives:
+
+{script}
+
+For each issue found:
+1. Identify the risky command or pattern
+2. Explain the risk
+3. Provide a safer alternative
+4. Suggest improvement
+
+Format your response:"""
+
+    SHELL_TYPES = ["bash", "zsh", "sh"]
+
+    @classmethod
+    def get_generate_prompt(
+        cls, description: str, shell_type: str = "bash"
+    ) -> str:
+        """Get generation prompt for shell type."""
+        return cls.BASH_GENERATE.format(
+            description=description, shell=shell_type
+        )
+
+    @classmethod
+    def get_explain_prompt(cls, script: str, shell_type: str = "bash") -> str:
+        """Get explanation prompt for script."""
+        return cls.BASH_EXPLAIN.format(script=script, shell=shell_type)
+
+    @classmethod
+    def get_refactor_prompt(cls, script: str, shell_type: str = "bash") -> str:
+        """Get refactor prompt for script."""
+        return cls.BASH_REFACTOR.format(script=script, shell=shell_type)
+
+
+class ShellParser:
+    """Parser for shell scripts."""
+
+    LINE_PATTERN = re.compile(r"^(?:(\s*)(.*?)(\s*))$", re.MULTILINE)
+    COMMENT_PATTERN = re.compile(r"^\s*#")
+    SHEBANG_PATTERN = re.compile(r"^#!(.+)$", re.MULTILINE)
+
+    @classmethod
+    def detect_shell(cls, script: str) -> str:
+        """Detect shell type from shebang.
+
+        Args:
+            script: Shell script content
+
+        Returns:
+            Detected shell type
+        """
+        shebang_match = cls.SHEBANG_PATTERN.match(script)
+        if shebang_match:
+            shebang = shebang_match.group(1)
+            if "zsh" in shebang:
+                return "zsh"
+            elif "bash" in shebang:
+                return "bash"
+            elif "sh" in shebang:
+                return "sh"
+        return "bash"
+
+    @classmethod
+    def parse_lines(cls, script: str) -> List[Tuple[int, str]]:
+        """Parse script into lines.
+
+        Args:
+            script: Shell script content
+
+        Returns:
+            List of (line_number, line_content) tuples
+        """
+        lines = script.split("\n")
+        return [(i + 1, line) for i, line in enumerate(lines)]
+
+    @classmethod
+    def extract_commands(cls, script: str) -> List[str]:
+        """Extract executable commands from script.
+
+        Args:
+            script: Shell script content
+
+        Returns:
+            List of commands
+        """
+        commands = []
+        for _, line in cls.parse_lines(script):
+            stripped = line.strip()
+            if stripped and not stripped.startswith("#") and not stripped.startswith(":"):
+                commands.append(stripped)
+        return commands
+
+
+class ShellGenerator:
+    """Shell generation engine."""
+
+    def __init__(self):
+        """Initialize generator."""
+        self.client = get_ollama_client()
+        self.parser = ShellParser()
+        self.templates = PromptTemplates
+
+    def generate(
+        self,
+        description: str,
+        shell_type: str = "bash",
+        model: Optional[str] = None,
+    ) -> GeneratedScript:
+        """Generate shell script from description.
+
+        Args:
+            description: Natural language description
+            shell_type: Target shell type
+            model: Ollama model to use
+
+        Returns:
+            GeneratedScript with commands and explanation
+        """
+        prompt = self.templates.get_generate_prompt(description, shell_type)
+
+        result = self.client.generate(prompt, model=model)
+
+        if result["success"]:
+            raw = result["response"].get("response", "")
+            return self._parse_generation(raw, shell_type)
+        else:
+            return GeneratedScript(
+                commands=[],
+                explanation=f"Generation failed: {result.get('error', 'Unknown error')}",
+                shell_type=shell_type,
+                raw_response="",
+            )
+
+    def _parse_generation(self, raw: str, shell_type: str) -> GeneratedScript:
+        """Parse raw LLM response.
+
+        Args:
+            raw: Raw response text
+            shell_type: Shell type
+
+        Returns:
+            GeneratedScript
+        """
+        commands = self._extract_commands(raw)
+        explanation = self._generate_summary(commands, raw)
+        return GeneratedScript(
+            commands=commands,
+            explanation=explanation,
+            shell_type=shell_type,
+            raw_response=raw,
+        )
+
+    def _extract_commands(self, raw: str) -> List[str]:
+        """Extract commands from raw response.
+
+        Args:
+            raw: Raw LLM response
+
+        Returns:
+            List of commands
+        """
+        lines = raw.strip().split("\n")
+        commands = []
+        for line in lines:
+            line = line.strip()
+            if line and not line.startswith("#"):
+                commands.append(line)
+        return commands
+
+    def _generate_summary(
+        self, commands: List[str], raw: str
+    ) -> str:
+        """Generate explanation summary.
+
+        Args:
+            commands: List of commands
+            raw: Raw response
+
+        Returns:
+            Explanation summary
+        """
+        return f"Generated {len(commands)} command(s) for your request."
+
+
+class ShellSafetyChecker:
+    """Safety checker for shell commands."""
+
+    DANGEROUS_PATTERNS = [
+        (r"rm\s+-rf\s+/", "Removes all files recursively - catastrophic data loss"),
+        (r":\(\)\s*\{\s*:\s*\|\s*:\s&\s*;\s*\}", "Fork bomb - can crash system"),
+        (r"chmod\s+777", "Removes all file permissions - security risk"),
+        (r"sudo\s+su", "Escalates to root without proper controls"),
+        (r"dd\s+if=/dev/zero", "Can overwrite disks if misconfigured"),
+        (r">\s*/dev/sda", "Direct disk write - destructive"),
+        (r"mv\s+.*\s+/dev/null", "Sends data to null device - data loss"),
+        (r"cat\s+.*\|\s*sh", "Executes arbitrary shell code"),
+        (r"curl.*\|\s*sh", "Downloads and executes code - security risk"),
+        (r"wget.*\|\s*sh", "Downloads and executes code - security risk"),
+    ]
+
+    WARNING_PATTERNS = [
+        (r"rm\s+-r", "Recursive removal"),
+        (r"rm\s+[^-]", "File removal"),
+        (r"chmod\s+[0-7][0-7][0-7]", "Permission change"),
+        (r"sudo", "Requires sudo privileges"),
+        (r">\s*\S+", "File redirection"),
+        (r"\|\s*bash", "Pipe to shell"),
+    ]
+
+    def __init__(self, safety_level: str = "moderate"):
+        """Initialize safety checker.
+
+        Args:
+            safety_level: strict, moderate, or permissive
+        """
+        self.safety_level = safety_level
+        self.config = get_config()
+
+    def check_command(self, command: str) -> Tuple[bool, List[str]]:
+        """Check if command is dangerous.
+
+        Args:
+            command: Shell command to check
+
+        Returns:
+            Tuple of (is_safe, list of warnings)
+        """
+        warnings = []
+        is_safe = True
+
+        for pattern, message in self.DANGEROUS_PATTERNS:
+            if re.search(pattern, command):
+                if self.safety_level in ["strict", "moderate"]:
+                    is_safe = False
+                warnings.append(f"DANGEROUS: {message}")
+
+        for pattern, message in self.WARNING_PATTERNS:
+            if re.search(pattern, command):
+                if self.safety_level == "strict":
+                    is_safe = False
+                warnings.append(f"WARNING: {message}")
+
+        if self.safety_level == "strict" and not warnings:
+            is_safe = True
+
+        return is_safe, warnings
+
+    def check_script(self, script: str) -> Dict[str, Any]:
+        """Check entire script for safety issues.
+
+        Args:
+            script: Shell script content
+
+        Returns:
+            Dictionary with safety assessment
+        """
+        lines = script.split("\n")
+        issues = []
+        warnings = []
+        safe_lines = []
+
+        for i, line in enumerate(lines, 1):
+            stripped = line.strip()
+            if stripped and not stripped.startswith("#"):
+                is_safe, line_warnings = self.check_command(stripped)
+                if not is_safe:
+                    issues.append({"line": i, "command": stripped, "warnings": line_warnings})
+                else:
+                    safe_lines.append(i)
+                warnings.extend(line_warnings)
+
+        return {
+            "is_safe": len(issues) == 0,
+            "issues": issues,
+            "warnings": warnings,
+            "safe_lines": safe_lines,
+            "total_lines": len([l for l in lines if l.strip() and not l.strip().startswith("#")]),
+        }
+
+
+def generate_shell(
+    description: str,
+    shell_type: str = "bash",
+    safety_level: Optional[str] = None,
+) -> GeneratedScript:
+    """Convenience function to generate shell script.
+
+    Args:
+        description: Natural language description
+        shell_type: Target shell type
+        safety_level: Safety level override
+
+    Returns:
+        GeneratedScript
+    """
+    generator = ShellGenerator()
+    if safety_level:
+        config = get_config()
+        config.config["safety"]["level"] = safety_level
+    return generator.generate(description, shell_type)