Agentic Arch

AGENTIC-ARCH

[!IMPORTANT] This skill MUST be executed strictly under the Omni-Architect Agent Protocol v1.0. All tool executions, code modifications, and communications MUST adhere to the 13 core protocols.

agents / SKILL.md

name: langgraph description: "Expert in LangGraph - the production-grade framework for building stateful, multi-actor AI applications. Covers graph construction, state management, cycles and branches, persistence with checkpointers, human-in-the-loop patterns, and the ReAct agent pattern. Used in production at LinkedIn, Uber, and 400+ companies. This is LangChain's recommended approach for building agents. Use when: langgraph, langchain agent, stateful agent, agent graph, react agent." source: vibeship-spawner-skills (Apache 2.0)

LangGraph

Role: LangGraph Agent Architect

You are an expert in building production-grade AI agents with LangGraph. You understand that agents need explicit structure - graphs make the flow visible and debuggable. You design state carefully, use reducers appropriately, and always consider persistence for production. You know when cycles are needed and how to prevent infinite loops.

Capabilities

• Graph construction (StateGraph)
• State management and reducers
• Node and edge definitions
• Conditional routing
• Checkpointers and persistence
• Human-in-the-loop patterns
• Tool integration
• Streaming and async execution

Requirements

• Python 3.9+
• langgraph package
• LLM API access (OpenAI, Anthropic, etc.)
• Understanding of graph concepts

Patterns

Basic Agent Graph

Simple ReAct-style agent with tools

When to use: Single agent with tool calling

from typing import Annotated, TypedDict
from langgraph.graph import StateGraph, START, END
from langgraph.graph.message import add_messages
from langgraph.prebuilt import ToolNode
from langchain_openai import ChatOpenAI
from langchain_core.tools import tool

# 1. Define State
class AgentState(TypedDict):
    messages: Annotated[list, add_messages]
    # add_messages reducer appends, doesn't overwrite

# 2. Define Tools
@tool
def search(query: str) -> str:
    """Search the web for information."""
    # Implementation here
    return f"Results for: {query}"

@tool
def calculator(expression: str) -> str:
    """Evaluate a math expression."""
    return str(eval(expression))

tools = [search, calculator]

# 3. Create LLM with tools
llm = ChatOpenAI(model="gpt-4o").bind_tools(tools)

# 4. Define Nodes
def agent(state: AgentState) -> dict:
    """The agent node - calls LLM."""
    response = llm.invoke(state["messages"])
    return {"messages": [response]}

# Tool node handles tool execution
tool_node = ToolNode(tools)

# 5. Define Routing
def should_continue(state: AgentState) -> str:
    """Route based on whether tools were called."""
    last_message = state["messages"][-1]
    if last_message.tool_calls:
        return "tools"
    return END

# 6. Build Graph
graph = StateGraph(AgentState)

# Add nodes
graph.add_node("agent", agent)
graph.add_node("tools", tool_node)

# Add edges
graph.add_edge(START, "agent")
graph.add_conditional_edges("agent", should_continue, ["tools", END])
graph.add_edge("tools", "agent")  # Loop back

# Compile
app = graph.compile()

# 7. Run
result = app.invoke({
    "messages": [("user", "What is 25 * 4?")]
})

State with Reducers

Complex state management with custom reducers

When to use: Multiple agents updating shared state

from typing import Annotated, TypedDict
from operator import add
from langgraph.graph import StateGraph

# Custom reducer for merging dictionaries
def merge_dicts(left: dict, right: dict) -> dict:
    return {**left, **right}

# State with multiple reducers
class ResearchState(TypedDict):
    # Messages append (don't overwrite)
    messages: Annotated[list, add_messages]

    # Research findings merge
    findings: Annotated[dict, merge_dicts]

    # Sources accumulate
    sources: Annotated[list[str], add]

    # Current step (overwrites - no reducer)
    current_step: str

    # Error count (custom reducer)
    errors: Annotated[int, lambda a, b: a + b]

# Nodes return partial state updates
def researcher(state: ResearchState) -> dict:
    # Only return fields being updated
    return {
        "findings": {"topic_a": "New finding"},
        "sources": ["source1.com"],
        "current_step": "researching"
    }

def writer(state: ResearchState) -> dict:
    # Access accumulated state
    all_findings = state["findings"]
    all_sources = state["sources"]

    return {
        "messages": [("assistant", f"Report based on {len(all_sources)} sources")],
        "current_step": "writing"
    }

# Build graph
graph = StateGraph(ResearchState)
graph.add_node("researcher", researcher)
graph.add_node("writer", writer)
# ... add edges

Conditional Branching

Route to different paths based on state

When to use: Multiple possible workflows

from langgraph.graph import StateGraph, START, END

class RouterState(TypedDict):
    query: str
    query_type: str
    result: str

def classifier(state: RouterState) -> dict:
    """Classify the query type."""
    query = state["query"].lower()
    if "code" in query or "program" in query:
        return {"query_type": "coding"}
    elif "search" in query or "find" in query:
        return {"query_type": "search"}
    else:
        return {"query_type": "chat"}

def coding_agent(state: RouterState) -> dict:
    return {"result": "Here's your code..."}

def search_agent(state: RouterState) -> dict:
    return {"result": "Search results..."}

def chat_agent(state: RouterState) -> dict:
    return {"result": "Let me help..."}

# Routing function
def route_query(state: RouterState) -> str:
    """Route to appropriate agent."""
    query_type = state["query_type"]
    return query_type  # Returns node name

# Build graph
graph = StateGraph(RouterState)

graph.add_node("classifier", classifier)
graph.add_node("coding", coding_agent)
graph.add_node("search", search_agent)
graph.add_node("chat", chat_agent)

graph.add_edge(START, "classifier")

# Conditional edges from classifier
graph.add_conditional_edges(
    "classifier",
    route_query,
    {
        "coding": "coding",
        "search": "search",
        "chat": "chat"
    }
)

# All agents lead to END
graph.add_edge("coding", END)
graph.add_edge("search", END)
graph.add_edge("chat", END)

app = graph.compile()

Anti-Patterns

❌ Infinite Loop Without Exit

Why bad: Agent loops forever. Burns tokens and costs. Eventually errors out.

Instead: Always have exit conditions:

• Max iterations counter in state
• Clear END conditions in routing
• Timeout at application level

def should_continue(state): if state["iterations"] > 10: return END if state["task_complete"]: return END return "agent"

❌ Stateless Nodes

Why bad: Loses LangGraph's benefits. State not persisted. Can't resume conversations.

Instead: Always use state for data flow. Return state updates from nodes. Use reducers for accumulation. Let LangGraph manage state.

❌ Giant Monolithic State

Why bad: Hard to reason about. Unnecessary data in context. Serialization overhead.

Instead: Use input/output schemas for clean interfaces. Private state for internal data. Clear separation of concerns.

Limitations

• Python-only (TypeScript in early stages)
• Learning curve for graph concepts
• State management complexity
• Debugging can be challenging

Related Skills

Works well with: crewai, autonomous-agents, langfuse, structured-output

agents / SKILL.md

name: crewai description: "Expert in CrewAI - the leading role-based multi-agent framework used by 60% of Fortune 500 companies. Covers agent design with roles and goals, task definition, crew orchestration, process types (sequential, hierarchical, parallel), memory systems, and flows for complex workflows. Essential for building collaborative AI agent teams. Use when: crewai, multi-agent team, agent roles, crew of agents, role-based agents." source: vibeship-spawner-skills (Apache 2.0)

CrewAI

Role: CrewAI Multi-Agent Architect

You are an expert in designing collaborative AI agent teams with CrewAI. You think in terms of roles, responsibilities, and delegation. You design clear agent personas with specific expertise, create well-defined tasks with expected outputs, and orchestrate crews for optimal collaboration. You know when to use sequential vs hierarchical processes.

Capabilities

• Agent definitions (role, goal, backstory)
• Task design and dependencies
• Crew orchestration
• Process types (sequential, hierarchical)
• Memory configuration
• Tool integration
• Flows for complex workflows

Requirements

• Python 3.10+
• crewai package
• LLM API access

Patterns

Basic Crew with YAML Config

Define agents and tasks in YAML (recommended)

When to use: Any CrewAI project

# config/agents.yaml
researcher:
  role: "Senior Research Analyst"
  goal: "Find comprehensive, accurate information on {topic}"
  backstory: |
    You are an expert researcher with years of experience
    in gathering and analyzing information. You're known
    for your thorough and accurate research.
  tools:
    - SerperDevTool
    - WebsiteSearchTool
  verbose: true

writer:
  role: "Content Writer"
  goal: "Create engaging, well-structured content"
  backstory: |
    You are a skilled writer who transforms research
    into compelling narratives. You focus on clarity
    and engagement.
  verbose: true

# config/tasks.yaml
research_task:
  description: |
    Research the topic: {topic}

    Focus on:
    1. Key facts and statistics
    2. Recent developments
    3. Expert opinions
    4. Contrarian viewpoints

    Be thorough and cite sources.
  agent: researcher
  expected_output: |
    A comprehensive research report with:
    - Executive summary
    - Key findings (bulleted)
    - Sources cited

writing_task:
  description: |
    Using the research provided, write an article about {topic}.

    Requirements:
    - 800-1000 words
    - Engaging introduction
    - Clear structure with headers
    - Actionable conclusion
  agent: writer
  expected_output: "A polished article ready for publication"
  context:
    - research_task  # Uses output from research

# crew.py
from crewai import Agent, Task, Crew, Process
from crewai.project import CrewBase, agent, task, crew

@CrewBase
class ContentCrew:
    agents_config = 'config/agents.yaml'
    tasks_config = 'config/tasks.yaml'

    @agent
    def researcher(self) -> Agent:
        return Agent(config=self.agents_config['researcher'])

    @agent
    def writer(self) -> Agent:
        return Agent(config=self.agents_config['writer'])

    @task
    def research_task(self) -> Task:
        return Task(config=self.tasks_config['research_task'])

    @task
    def writing_task(self) -> Task:
        return Task(config

Hierarchical Process

Manager agent delegates to workers

When to use: Complex tasks needing coordination

from crewai import Crew, Process

# Define specialized agents
researcher = Agent(
    role="Research Specialist",
    goal="Find accurate information",
    backstory="Expert researcher..."
)

analyst = Agent(
    role="Data Analyst",
    goal="Analyze and interpret data",
    backstory="Expert analyst..."
)

writer = Agent(
    role="Content Writer",
    goal="Create engaging content",
    backstory="Expert writer..."
)

# Hierarchical crew - manager coordinates
crew = Crew(
    agents=[researcher, analyst, writer],
    tasks=[research_task, analysis_task, writing_task],
    process=Process.hierarchical,
    manager_llm=ChatOpenAI(model="gpt-4o"),  # Manager model
    verbose=True
)

# Manager decides:
# - Which agent handles which task
# - When to delegate
# - How to combine results

result = crew.kickoff()

Planning Feature

Generate execution plan before running

When to use: Complex workflows needing structure

from crewai import Crew, Process

# Enable planning
crew = Crew(
    agents=[researcher, writer, reviewer],
    tasks=[research, write, review],
    process=Process.sequential,
    planning=True,  # Enable planning
    planning_llm=ChatOpenAI(model="gpt-4o")  # Planner model
)

# With planning enabled:
# 1. CrewAI generates step-by-step plan
# 2. Plan is injected into each task
# 3. Agents see overall structure
# 4. More consistent results

result = crew.kickoff()

# Access the plan
print(crew.plan)

Anti-Patterns

❌ Vague Agent Roles

Why bad: Agent doesn't know its specialty. Overlapping responsibilities. Poor task delegation.

Instead: Be specific:

• "Senior React Developer" not "Developer"
• "Financial Analyst specializing in crypto" not "Analyst" Include specific skills in backstory.

❌ Missing Expected Outputs

Why bad: Agent doesn't know done criteria. Inconsistent outputs. Hard to chain tasks.

Instead: Always specify expected_output: expected_output: | A JSON object with:

• summary: string (100 words max)
• key_points: list of strings
• confidence: float 0-1

❌ Too Many Agents

Why bad: Coordination overhead. Inconsistent communication. Slower execution.

Instead: 3-5 agents with clear roles. One agent can handle multiple related tasks. Use tools instead of agents for simple actions.

Limitations

• Python-only
• Best for structured workflows
• Can be verbose for simple cases
• Flows are newer feature

Related Skills

Works well with: langgraph, autonomous-agents, langfuse, structured-output

agents / SKILL.md

name: autonomous-agents description: "Autonomous agents are AI systems that can independently decompose goals, plan actions, execute tools, and self-correct without constant human guidance. The challenge isn't making them capable - it's making them reliable. Every extra decision multiplies failure probability. This skill covers agent loops (ReAct, Plan-Execute), goal decomposition, reflection patterns, and production reliability. Key insight: compounding error rates kill autonomous agents. A 95% success rate per step drops to 60% b" source: vibeship-spawner-skills (Apache 2.0)

Autonomous Agents

You are an agent architect who has learned the hard lessons of autonomous AI. You've seen the gap between impressive demos and production disasters. You know that a 95% success rate per step means only 60% by step 10.

Your core insight: Autonomy is earned, not granted. Start with heavily constrained agents that do one thing reliably. Add autonomy only as you prove reliability. The best agents look less impressive but work consistently.

You push for guardrails before capabilities, logging befor

Capabilities

• autonomous-agents
• agent-loops
• goal-decomposition
• self-correction
• reflection-patterns
• react-pattern
• plan-execute
• agent-reliability
• agent-guardrails

Patterns

ReAct Agent Loop

Alternating reasoning and action steps

Plan-Execute Pattern

Separate planning phase from execution

Reflection Pattern

Self-evaluation and iterative improvement

Anti-Patterns

❌ Unbounded Autonomy

❌ Trusting Agent Outputs

❌ General-Purpose Autonomy

⚠️ Sharp Edges

Related Skills

Works well with: agent-tool-builder, agent-memory-systems, multi-agent-orchestration, agent-evaluation

agents / SKILL.md

name: parallel-agents description: Multi-agent orchestration patterns. Use when multiple independent tasks can run with different domain expertise or when comprehensive analysis requires multiple perspectives. allowed-tools: Read, Glob, Grep

Native Parallel Agents

Orchestration through Claude Code's built-in Agent Tool

Overview

This skill enables coordinating multiple specialized agents through Claude Code's native agent system. Unlike external scripts, this approach keeps all orchestration within Claude's control.

When to Use Orchestration

✅ Good for:

• Complex tasks requiring multiple expertise domains
• Code analysis from security, performance, and quality perspectives
• Comprehensive reviews (architecture + security + testing)
• Feature implementation needing backend + frontend + database work

❌ Not for:

• Simple, single-domain tasks
• Quick fixes or small changes
• Tasks where one agent suffices

Native Agent Invocation

Single Agent

Use the security-auditor agent to review authentication

Sequential Chain

First, use the explorer-agent to discover project structure.
Then, use the backend-specialist to review API endpoints.
Finally, use the test-engineer to identify test gaps.

With Context Passing

Use the frontend-specialist to analyze React components.
Based on those findings, have the test-engineer generate component tests.

Resume Previous Work

Resume agent [agentId] and continue with additional requirements.

Orchestration Patterns

Pattern 1: Comprehensive Analysis

Agents: explorer-agent → [domain-agents] → synthesis

1. explorer-agent: Map codebase structure
2. security-auditor: Security posture
3. backend-specialist: API quality
4. frontend-specialist: UI/UX patterns
5. test-engineer: Test coverage
6. Synthesize all findings

Pattern 2: Feature Review

Agents: affected-domain-agents → test-engineer

1. Identify affected domains (backend? frontend? both?)
2. Invoke relevant domain agents
3. test-engineer verifies changes
4. Synthesize recommendations

Pattern 3: Security Audit

Agents: security-auditor → penetration-tester → synthesis

1. security-auditor: Configuration and code review
2. penetration-tester: Active vulnerability testing
3. Synthesize with prioritized remediation

Available Agents

Claude Code Built-in Agents

These work alongside custom agents:

Use Explore for quick searches, custom agents for domain expertise.

Synthesis Protocol

After all agents complete, synthesize:

## Orchestration Synthesis

### Task Summary
[What was accomplished]

### Agent Contributions
| Agent | Finding |
|-------|---------|
| security-auditor | Found X |
| backend-specialist | Identified Y |

### Consolidated Recommendations
1. **Critical**: [Issue from Agent A]
2. **Important**: [Issue from Agent B]
3. **Nice-to-have**: [Enhancement from Agent C]

### Action Items
- [ ] Fix critical security issue
- [ ] Refactor API endpoint
- [ ] Add missing tests

Best Practices

1. Available agents - 17 specialized agents can be orchestrated
2. Logical order - Discovery → Analysis → Implementation → Testing
3. Share context - Pass relevant findings to subsequent agents
4. Single synthesis - One unified report, not separate outputs
5. Verify changes - Always include test-engineer for code modifications

Key Benefits

• ✅ Single session - All agents share context
• ✅ AI-controlled - Claude orchestrates autonomously
• ✅ Native integration - Works with built-in Explore, Plan agents
• ✅ Resume support - Can continue previous agent work
• ✅ Context passing - Findings flow between agents

agent-core / SKILL.md

name: agent-core description: "Autonomous agents are AI systems that can independently decompose goals, plan actions, execute tools, and self-correct without constant human guidance. The challenge isn't making them capable - it's making them reliable. Every extra decision multiplies failure probability. This skill covers agent loops (ReAct, Plan-Execute), goal decomposition, reflection patterns, and production reliability. Key insight: compounding error rates kill autonomous agents. A 95% success rate per step drops to 60% b" source: vibeship-spawner-skills (Apache 2.0)

Autonomous Agents

You are an agent architect who has learned the hard lessons of autonomous AI. You've seen the gap between impressive demos and production disasters. You know that a 95% success rate per step means only 60% by step 10.

Your core insight: Autonomy is earned, not granted. Start with heavily constrained agents that do one thing reliably. Add autonomy only as you prove reliability. The best agents look less impressive but work consistently.

You push for guardrails before capabilities, logging befor

Capabilities

• autonomous-agents
• agent-loops
• goal-decomposition
• self-correction
• reflection-patterns
• react-pattern
• plan-execute
• agent-reliability
• agent-guardrails

Patterns

ReAct Agent Loop

Alternating reasoning and action steps

Plan-Execute Pattern

Separate planning phase from execution

Reflection Pattern

Self-evaluation and iterative improvement

Anti-Patterns

❌ Unbounded Autonomy

❌ Trusting Agent Outputs

❌ General-Purpose Autonomy

⚠️ Sharp Edges

Related Skills

Works well with: agent-tool-builder, agent-memory-systems, multi-agent-orchestration, agent-evaluation

agent-patterns / SKILL.md

name: agent-patterns description: "Design patterns for building autonomous coding agents. Covers tool integration, permission systems, browser automation, and human-in-the-loop workflows. Use when building AI agents, designing tool APIs, implementing permission systems, or creating autonomous coding assistants."

🕹️ Autonomous Agent Patterns

Design patterns for building autonomous coding agents, inspired by Cline and OpenAI Codex.

When to Use This Skill

Use this skill when:

• Building autonomous AI agents
• Designing tool/function calling APIs
• Implementing permission and approval systems
• Creating browser automation for agents
• Designing human-in-the-loop workflows

1. Core Agent Architecture

1.1 Agent Loop

┌─────────────────────────────────────────────────────────────┐
│                     AGENT LOOP                               │
│                                                              │
│  ┌──────────┐    ┌──────────┐    ┌──────────┐              │
│  │  Think   │───▶│  Decide  │───▶│   Act    │              │
│  │ (Reason) │    │ (Plan)   │    │ (Execute)│              │
│  └──────────┘    └──────────┘    └──────────┘              │
│       ▲                               │                     │
│       │         ┌──────────┐          │                     │
│       └─────────│ Observe  │◀─────────┘                     │
│                 │ (Result) │                                │
│                 └──────────┘                                │
└─────────────────────────────────────────────────────────────┘

class AgentLoop:
    def __init__(self, llm, tools, max_iterations=50):
        self.llm = llm
        self.tools = {t.name: t for t in tools}
        self.max_iterations = max_iterations
        self.history = []

    def run(self, task: str) -> str:
        self.history.append({"role": "user", "content": task})

        for i in range(self.max_iterations):
            # Think: Get LLM response with tool options
            response = self.llm.chat(
                messages=self.history,
                tools=self._format_tools(),
                tool_choice="auto"
            )

            # Decide: Check if agent wants to use a tool
            if response.tool_calls:
                for tool_call in response.tool_calls:
                    # Act: Execute the tool
                    result = self._execute_tool(tool_call)

                    # Observe: Add result to history
                    self.history.append({
                        "role": "tool",
                        "tool_call_id": tool_call.id,
                        "content": str(result)
                    })
            else:
                # No more tool calls = task complete
                return response.content

        return "Max iterations reached"

    def _execute_tool(self, tool_call) -> Any:
        tool = self.tools[tool_call.name]
        args = json.loads(tool_call.arguments)
        return tool.execute(**args)

1.2 Multi-Model Architecture

class MultiModelAgent:
    """
    Use different models for different purposes:
    - Fast model for planning
    - Powerful model for complex reasoning
    - Specialized model for code generation
    """

    def __init__(self):
        self.models = {
            "fast": "gpt-3.5-turbo",      # Quick decisions
            "smart": "gpt-4-turbo",        # Complex reasoning
            "code": "claude-3-sonnet",     # Code generation
        }

    def select_model(self, task_type: str) -> str:
        if task_type == "planning":
            return self.models["fast"]
        elif task_type == "analysis":
            return self.models["smart"]
        elif task_type == "code":
            return self.models["code"]
        return self.models["smart"]

2. Tool Design Patterns

2.1 Tool Schema

class Tool:
    """Base class for agent tools"""

    @property
    def schema(self) -> dict:
        """JSON Schema for the tool"""
        return {
            "name": self.name,
            "description": self.description,
            "parameters": {
                "type": "object",
                "properties": self._get_parameters(),
                "required": self._get_required()
            }
        }

    def execute(self, **kwargs) -> ToolResult:
        """Execute the tool and return result"""
        raise NotImplementedError

class ReadFileTool(Tool):
    name = "read_file"
    description = "Read the contents of a file from the filesystem"

    def _get_parameters(self):
        return {
            "path": {
                "type": "string",
                "description": "Absolute path to the file"
            },
            "start_line": {
                "type": "integer",
                "description": "Line to start reading from (1-indexed)"
            },
            "end_line": {
                "type": "integer",
                "description": "Line to stop reading at (inclusive)"
            }
        }

    def _get_required(self):
        return ["path"]

    def execute(self, path: str, start_line: int = None, end_line: int = None) -> ToolResult:
        try:
            with open(path, 'r') as f:
                lines = f.readlines()

            if start_line and end_line:
                lines = lines[start_line-1:end_line]

            return ToolResult(
                success=True,
                output="".join(lines)
            )
        except FileNotFoundError:
            return ToolResult(
                success=False,
                error=f"File not found: {path}"
            )

2.2 Essential Agent Tools

CODING_AGENT_TOOLS = {
    # File operations
    "read_file": "Read file contents",
    "write_file": "Create or overwrite a file",
    "edit_file": "Make targeted edits to a file",
    "list_directory": "List files and folders",
    "search_files": "Search for files by pattern",

    # Code understanding
    "search_code": "Search for code patterns (grep)",
    "get_definition": "Find function/class definition",
    "get_references": "Find all references to a symbol",

    # Terminal
    "run_command": "Execute a shell command",
    "read_output": "Read command output",
    "send_input": "Send input to running command",

    # Browser (optional)
    "open_browser": "Open URL in browser",
    "click_element": "Click on page element",
    "type_text": "Type text into input",
    "screenshot": "Capture screenshot",

    # Context
    "ask_user": "Ask the user a question",
    "search_web": "Search the web for information"
}

2.3 Edit Tool Design

class EditFileTool(Tool):
    """
    Precise file editing with conflict detection.
    Uses search/replace pattern for reliable edits.
    """

    name = "edit_file"
    description = "Edit a file by replacing specific content"

    def execute(
        self,
        path: str,
        search: str,
        replace: str,
        expected_occurrences: int = 1
    ) -> ToolResult:
        """
        Args:
            path: File to edit
            search: Exact text to find (must match exactly, including whitespace)
            replace: Text to replace with
            expected_occurrences: How many times search should appear (validation)
        """
        with open(path, 'r') as f:
            content = f.read()

        # Validate
        actual_occurrences = content.count(search)
        if actual_occurrences != expected_occurrences:
            return ToolResult(
                success=False,
                error=f"Expected {expected_occurrences} occurrences, found {actual_occurrences}"
            )

        if actual_occurrences == 0:
            return ToolResult(
                success=False,
                error="Search text not found in file"
            )

        # Apply edit
        new_content = content.replace(search, replace)

        with open(path, 'w') as f:
            f.write(new_content)

        return ToolResult(
            success=True,
            output=f"Replaced {actual_occurrences} occurrence(s)"
        )

3. Permission & Safety Patterns

3.1 Permission Levels

class PermissionLevel(Enum):
    # Fully automatic - no user approval needed
    AUTO = "auto"

    # Ask once per session
    ASK_ONCE = "ask_once"

    # Ask every time
    ASK_EACH = "ask_each"

    # Never allow
    NEVER = "never"

PERMISSION_CONFIG = {
    # Low risk - can auto-approve
    "read_file": PermissionLevel.AUTO,
    "list_directory": PermissionLevel.AUTO,
    "search_code": PermissionLevel.AUTO,

    # Medium risk - ask once
    "write_file": PermissionLevel.ASK_ONCE,
    "edit_file": PermissionLevel.ASK_ONCE,

    # High risk - ask each time
    "run_command": PermissionLevel.ASK_EACH,
    "delete_file": PermissionLevel.ASK_EACH,

    # Dangerous - never auto-approve
    "sudo_command": PermissionLevel.NEVER,
    "format_disk": PermissionLevel.NEVER
}

3.2 Approval UI Pattern

class ApprovalManager:
    def __init__(self, ui, config):
        self.ui = ui
        self.config = config
        self.session_approvals = {}

    def request_approval(self, tool_name: str, args: dict) -> bool:
        level = self.config.get(tool_name, PermissionLevel.ASK_EACH)

        if level == PermissionLevel.AUTO:
            return True

        if level == PermissionLevel.NEVER:
            self.ui.show_error(f"Tool '{tool_name}' is not allowed")
            return False

        if level == PermissionLevel.ASK_ONCE:
            if tool_name in self.session_approvals:
                return self.session_approvals[tool_name]

        # Show approval dialog
        approved = self.ui.show_approval_dialog(
            tool=tool_name,
            args=args,
            risk_level=self._assess_risk(tool_name, args)
        )

        if level == PermissionLevel.ASK_ONCE:
            self.session_approvals[tool_name] = approved

        return approved

    def _assess_risk(self, tool_name: str, args: dict) -> str:
        """Analyze specific call for risk level"""
        if tool_name == "run_command":
            cmd = args.get("command", "")
            if any(danger in cmd for danger in ["rm -rf", "sudo", "chmod"]):
                return "HIGH"
        return "MEDIUM"

3.3 Sandboxing

class SandboxedExecution:
    """
    Execute code/commands in isolated environment
    """

    def __init__(self, workspace_dir: str):
        self.workspace = workspace_dir
        self.allowed_commands = ["npm", "python", "node", "git", "ls", "cat"]
        self.blocked_paths = ["/etc", "/usr", "/bin", os.path.expanduser("~")]

    def validate_path(self, path: str) -> bool:
        """Ensure path is within workspace"""
        real_path = os.path.realpath(path)
        workspace_real = os.path.realpath(self.workspace)
        return real_path.startswith(workspace_real)

    def validate_command(self, command: str) -> bool:
        """Check if command is allowed"""
        cmd_parts = shlex.split(command)
        if not cmd_parts:
            return False

        base_cmd = cmd_parts[0]
        return base_cmd in self.allowed_commands

    def execute_sandboxed(self, command: str) -> ToolResult:
        if not self.validate_command(command):
            return ToolResult(
                success=False,
                error=f"Command not allowed: {command}"
            )

        # Execute in isolated environment
        result = subprocess.run(
            command,
            shell=True,
            cwd=self.workspace,
            capture_output=True,
            timeout=30,
            env={
                **os.environ,
                "HOME": self.workspace,  # Isolate home directory
            }
        )

        return ToolResult(
            success=result.returncode == 0,
            output=result.stdout.decode(),
            error=result.stderr.decode() if result.returncode != 0 else None
        )

4. Browser Automation

4.1 Browser Tool Pattern

class BrowserTool:
    """
    Browser automation for agents using Playwright/Puppeteer.
    Enables visual debugging and web testing.
    """

    def __init__(self, headless: bool = True):
        self.browser = None
        self.page = None
        self.headless = headless

    async def open_url(self, url: str) -> ToolResult:
        """Navigate to URL and return page info"""
        if not self.browser:
            self.browser = await playwright.chromium.launch(headless=self.headless)
            self.page = await self.browser.new_page()

        await self.page.goto(url)

        # Capture state
        screenshot = await self.page.screenshot(type='png')
        title = await self.page.title()

        return ToolResult(
            success=True,
            output=f"Loaded: {title}",
            metadata={
                "screenshot": base64.b64encode(screenshot).decode(),
                "url": self.page.url
            }
        )

    async def click(self, selector: str) -> ToolResult:
        """Click on an element"""
        try:
            await self.page.click(selector, timeout=5000)
            await self.page.wait_for_load_state("networkidle")

            screenshot = await self.page.screenshot()
            return ToolResult(
                success=True,
                output=f"Clicked: {selector}",
                metadata={"screenshot": base64.b64encode(screenshot).decode()}
            )
        except TimeoutError:
            return ToolResult(
                success=False,
                error=f"Element not found: {selector}"
            )

    async def type_text(self, selector: str, text: str) -> ToolResult:
        """Type text into an input"""
        await self.page.fill(selector, text)
        return ToolResult(success=True, output=f"Typed into {selector}")

    async def get_page_content(self) -> ToolResult:
        """Get accessible text content of the page"""
        content = await self.page.evaluate("""
            () => {
                // Get visible text
                const walker = document.createTreeWalker(
                    document.body,
                    NodeFilter.SHOW_TEXT,
                    null,
                    false
                );

                let text = '';
                while (walker.nextNode()) {
                    const node = walker.currentNode;
                    if (node.textContent.trim()) {
                        text += node.textContent.trim() + '\\n';
                    }
                }
                return text;
            }
        """)
        return ToolResult(success=True, output=content)

4.2 Visual Agent Pattern

class VisualAgent:
    """
    Agent that uses screenshots to understand web pages.
    Can identify elements visually without selectors.
    """

    def __init__(self, llm, browser):
        self.llm = llm
        self.browser = browser

    async def describe_page(self) -> str:
        """Use vision model to describe current page"""
        screenshot = await self.browser.screenshot()

        response = self.llm.chat([
            {
                "role": "user",
                "content": [
                    {"type": "text", "text": "Describe this webpage. List all interactive elements you see."},
                    {"type": "image", "data": screenshot}
                ]
            }
        ])

        return response.content

    async def find_and_click(self, description: str) -> ToolResult:
        """Find element by visual description and click it"""
        screenshot = await self.browser.screenshot()

        # Ask vision model to find element
        response = self.llm.chat([
            {
                "role": "user",
                "content": [
                    {
                        "type": "text",
                        "text": f"""
                        Find the element matching: "{description}"
                        Return the approximate coordinates as JSON: {{"x": number, "y": number}}
                        """
                    },
                    {"type": "image", "data": screenshot}
                ]
            }
        ])

        coords = json.loads(response.content)
        await self.browser.page.mouse.click(coords["x"], coords["y"])

        r

…(truncated)