diff --git a/crates/agent-protocol/src/protocol.rs b/crates/agent-protocol/src/protocol.rs
index f41f03d..6b5f07e 100644
--- a/crates/agent-protocol/src/protocol.rs
+++ b/crates/agent-protocol/src/protocol.rs
@@ -35,34 +35,85 @@ pub enum EventType {
     GuardrailInspect,
 }
 
-/// Agent decision
+/// Agent response decision indicating how to handle a request or response.
+///
+/// This enum represents the decision an agent makes when processing a request or response.
+/// It allows agents to allow, block, redirect, or challenge requests based on their processing logic.
+///
+/// # Variants
+///
+/// - **Allow**: Continue normal processing without modification
+/// - **Block**: Reject the request/response with a custom error response
+/// - **Redirect**: Send the client to a different URL
+/// - **Challenge**: Request additional verification from the client
+///
+/// # Examples
+///
+/// ```rust
+/// use std::collections::HashMap;
+/// use zentinel_agent_protocol::Decision;
+///
+/// // Allow request to proceed normally
+/// let allow = Decision::Allow;
+///
+/// // Block with 403 and custom body
+/// let block = Decision::Block {
+///     status: 403,
+///     body: Some("Access denied".to_string()),
+///     headers: None,
+/// };
+///
+/// // Redirect to login page
+/// let redirect = Decision::Redirect {
+///     url: "https://example.com/login".to_string(),
+///     status: 302,
+/// };
+///
+/// // Challenge with CAPTCHA
+/// let mut params = HashMap::new();
+/// params.insert("type".to_string(), "recaptcha".to_string());
+/// let challenge = Decision::Challenge {
+///     challenge_type: "captcha".to_string(),
+///     params,
+/// };
+/// ```
 #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Default)]
 #[serde(rename_all = "snake_case")]
 pub enum Decision {
-    /// Allow the request/response to continue
+    /// Allow the request/response to continue without modification.
+    ///
+    /// This is the default decision that indicates normal processing should proceed.
     #[default]
     Allow,
-    /// Block the request/response
+    /// Block the request/response with a custom error response.
+    ///
+    /// The proxy will return the specified status code and optional body/headers
+    /// instead of forwarding the request to the upstream or returning the original response.
     Block {
-        /// HTTP status code to return
+        /// HTTP status code to return (typically 4xx or 5xx)
         status: u16,
-        /// Optional response body
+        /// Optional response body content
         body: Option<String>,
-        /// Optional response headers
+        /// Optional response headers to include
         headers: Option<HashMap<String, String>>,
     },
-    /// Redirect the request
+    /// Redirect the client to a different URL.
+    ///
+    /// The proxy will return a redirect response with the specified URL and status code.
     Redirect {
-        /// Redirect URL
+        /// Target URL for the redirect
         url: String,
-        /// HTTP status code (301, 302, 303, 307, 308)
+        /// HTTP redirect status code (301, 302, 303, 307, or 308)
         status: u16,
     },
-    /// Challenge the client (e.g., CAPTCHA)
+    /// Request additional verification from the client.
+    ///
+    /// This can be used to implement CAPTCHA, multi-factor authentication,
+    /// or other challenge-response mechanisms.
     Challenge {
-        /// Challenge type
+        /// Type of challenge (e.g., "captcha", "otp", "totp")
         challenge_type: String,
-        /// Challenge parameters
+        /// Challenge-specific parameters and configuration
         params: HashMap<String, String>,
     },
 }
diff --git a/crates/agent-protocol/src/v2/pool.rs b/crates/agent-protocol/src/v2/pool.rs
index d2327cf..d61c79c 100644
--- a/crates/agent-protocol/src/v2/pool.rs
+++ b/crates/agent-protocol/src/v2/pool.rs
@@ -519,16 +519,50 @@ impl AgentEntry {
     }
 }
 
-/// Agent connection pool.
+/// Agent connection pool for managing multiple connections to external processing agents.
 ///
-/// Manages multiple connections to multiple agents with load balancing,
-/// health tracking, automatic reconnection, and metrics collection.
+/// `AgentPool` provides production-ready connection pooling, load balancing, health tracking,
+/// automatic reconnection, and metrics collection for robust agent communication.
+///
+/// # Features
+///
+/// - **Connection pooling**: Maintains multiple connections per agent for better throughput
+/// - **Load balancing**: Routes requests using round-robin, least-connections, or health-based strategies
+/// - **Health monitoring**: Tracks agent health and routes requests to healthy connections
+/// - **Automatic reconnection**: Reconnects failed connections transparently
+/// - **Flow control**: Manages backpressure and request queuing
+/// - **Metrics**: Collects detailed metrics on performance and health
+/// - **Configuration management**: Distributes config updates to agents
 ///
 /// # Performance
 ///
 /// Uses `DashMap` for lock-free reads in the hot path. Agent lookup is O(1)
 /// without contention. Connection selection uses cached health state to avoid
 /// async I/O per request.
+///
+/// # Example
+///
+/// ```rust
+/// use std::time::Duration;
+/// use zentinel_agent_protocol::v2::{AgentPool, AgentPoolConfig, LoadBalanceStrategy};
+/// use zentinel_agent_protocol::{RequestHeadersEvent};
+///
+/// // Create pool with custom config
+/// let config = AgentPoolConfig {
+///     connections_per_agent: 4,
+///     load_balance_strategy: LoadBalanceStrategy::LeastConnections,
+///     health_check_interval: Duration::from_secs(30),
+///     ..Default::default()
+/// };
+/// let pool = AgentPool::with_config(config);
+///
+/// // Add agent with Unix domain socket endpoint
+/// pool.add_agent("waf", "unix:/tmp/waf.sock").await?;
+///
+/// // Send request headers to an agent
+/// let headers = RequestHeadersEvent { /* ... */ };
+/// let response = pool.send_request_headers("waf", "correlation-123", &headers).await?;
+/// ```
 pub struct AgentPool {
     config: AgentPoolConfig,
     /// Lock-free concurrent map for agent lookup.
diff --git a/crates/agent-protocol/src/v2/server.rs b/crates/agent-protocol/src/v2/server.rs
index 4022310..3538a38 100644
--- a/crates/agent-protocol/src/v2/server.rs
+++ b/crates/agent-protocol/src/v2/server.rs
@@ -24,14 +24,63 @@ use crate::{
     WebSocketFrameEvent,
 };
 
-/// v2 agent handler trait.
+/// Trait for implementing agent handlers in Protocol v2.
 ///
-/// Implement this trait to create a v2 agent. The v2 handler adds:
-/// - Capability reporting
-/// - Health reporting
-/// - Flow control awareness
-/// - Metrics export
-/// - Configuration updates
+/// `AgentHandlerV2` defines the interface that agent implementations must provide
+/// to handle various types of events from the proxy. This includes request/response
+/// processing, WebSocket handling, health monitoring, and configuration management.
+///
+/// The trait provides sensible defaults for all methods, allowing agents to implement
+/// only the events they need to handle. All methods are async to support I/O operations.
+///
+/// # Features
+///
+/// - **Capability reporting**: Declare what the agent can process
+/// - **Health reporting**: Report current health status to the proxy
+/// - **Flow control awareness**: Handle backpressure and flow control
+/// - **Metrics export**: Provide metrics about agent performance
+/// - **Configuration updates**: Handle dynamic configuration changes
+///
+/// # Event Lifecycle
+///
+/// 1. **Handshake**: Agent declares capabilities when connecting
+/// 2. **Headers**: Process request/response headers first
+/// 3. **Body chunks**: Handle streaming body data if needed
+/// 4. **Completion**: Final processing when request/response is complete
+/// 5. **WebSocket**: Handle WebSocket frames for upgraded connections
+///
+/// # Example
+///
+/// ```rust
+/// use async_trait::async_trait;
+/// use zentinel_agent_protocol::v2::{AgentHandlerV2, AgentCapabilities, AgentResponse};
+/// use zentinel_agent_protocol::{EventType, RequestHeadersEvent};
+///
+/// pub struct MyWafAgent;
+///
+/// #[async_trait]
+/// impl AgentHandlerV2 for MyWafAgent {
+///     fn capabilities(&self) -> AgentCapabilities {
+///         AgentCapabilities::new("my-waf", "My WAF Agent", "1.0.0")
+///             .with_event(EventType::RequestHeaders)
+///     }
+///
+///     async fn on_request_headers(&self, event: RequestHeadersEvent) -> AgentResponse {
+///         // Inspect headers for malicious patterns
+///         if event.headers.contains_key("x-malicious") {
+///             AgentResponse::block(403, Some("Blocked by WAF".to_string()))
+///         } else {
+///             AgentResponse::default_allow()
+///         }
+///     }
+/// }
+/// ```
+///
+/// # Errors
+///
+/// Agent methods should return `AgentResponse` with appropriate `Decision` variants.
+/// Runtime errors should be logged internally rather than propagated, as the proxy
+/// needs to maintain high availability.
 #[async_trait]
 pub trait AgentHandlerV2: Send + Sync {
     /// Get agent capabilities.
@@ -121,7 +170,41 @@ pub enum DrainReason {
     Manual,
 }
 
-/// v2 gRPC agent server.
+/// gRPC-based agent server implementation for Protocol v2.
+///
+/// `GrpcAgentServerV2` provides a gRPC transport for agents that need to communicate
+/// with the Zentinel proxy over the network. This is ideal for agents running in
+/// separate processes, containers, or on different machines.
+///
+/// # Features
+///
+/// - **Network transport**: Communicates over TCP with HTTP/2 and TLS support
+/// - **Language agnostic**: Works with any gRPC client implementation
+/// - **Scalability**: Can handle multiple concurrent proxy connections
+/// - **Monitoring**: Integrates with gRPC ecosystem tools for observability
+///
+/// # Example
+///
+/// ```rust
+/// use zentinel_agent_protocol::v2::{GrpcAgentServerV2, AgentHandlerV2};
+///
+/// // Create server with your handler
+/// let handler = Box::new(MyAgent::new());
+/// let server = GrpcAgentServerV2::new("my-agent", handler);
+///
+/// // Serve on a specific address
+/// let addr = "127.0.0.1:8080".parse()?;
+/// server.run(addr).await?;
+/// ```
+///
+/// # Transport Details
+///
+/// The gRPC transport uses the standard Agent Protocol v2 service definition:
+/// - Bidirectional streaming for event processing
+/// - Capability negotiation during handshake
+/// - Health check integration
+/// - Configuration update support
+/// - Metrics collection
 pub struct GrpcAgentServerV2 {
     id: String,
     handler: Arc<dyn AgentHandlerV2>,
diff --git a/crates/agent-protocol/src/v2/uds_server.rs b/crates/agent-protocol/src/v2/uds_server.rs
index 853f756..fc8eb86 100644
--- a/crates/agent-protocol/src/v2/uds_server.rs
+++ b/crates/agent-protocol/src/v2/uds_server.rs
@@ -23,10 +23,53 @@ use crate::{
     RequestHeadersEvent, ResponseBodyChunkEvent, ResponseHeadersEvent, WebSocketFrameEvent,
 };
 
-/// v2 agent server over Unix Domain Socket.
+/// Unix Domain Socket agent server implementation for Protocol v2.
 ///
-/// Listens on a Unix socket, accepts connections, and dispatches events to an
-/// [`AgentHandlerV2`] implementation using the v2 binary wire format.
+/// `UdsAgentServerV2` provides a high-performance local transport for agents running
+/// on the same machine as the Zentinel proxy. It uses Unix Domain Sockets for
+/// inter-process communication with minimal overhead.
+///
+/// This transport is recommended for agents that need the lowest possible latency
+/// and are co-located with the proxy process.
+///
+/// # Features
+///
+/// - **High performance**: Minimal overhead for local communication
+/// - **Binary protocol**: Efficient binary encoding for reduced CPU usage
+/// - **File system permissions**: Uses Unix socket permissions for access control
+/// - **Socket cleanup**: Removes stale socket files on startup
+///
+/// # Example
+///
+/// ```rust
+/// use zentinel_agent_protocol::v2::{UdsAgentServerV2, AgentHandlerV2};
+/// use std::path::Path;
+///
+/// // Create server with your handler
+/// let handler = Box::new(MyAgent::new());
+/// let server = UdsAgentServerV2::new(
+///     "my-agent",
+///     Path::new("/var/run/my-agent.sock"),
+///     handler
+/// );
+///
+/// // Start listening
+/// server.run().await?;
+/// ```
+///
+/// # Socket Management
+///
+/// The server automatically:
+/// - Creates the socket file with appropriate permissions
+/// - Removes stale socket files on startup
+/// - Handles multiple concurrent proxy connections
+///
+/// # Errors
+///
+/// May return errors for:
+/// - Permission issues creating/accessing the socket file
+/// - Invalid socket paths or filesystem issues
+/// - Handler implementation errors
 pub struct UdsAgentServerV2 {
     id: String,
     socket_path: PathBuf,