Fork me on GitHub

Session Hooks

⚠️ Disclaimer: This is an unofficial, community-driven SDK and is not supported or endorsed by GitHub. Use at your own risk.

Session hooks allow you to intercept and modify tool execution, user prompts, and session lifecycle events. Use hooks to implement custom logic like logging, security controls, or context injection.


Overview

Hook When It's Called Can Modify
Pre-Tool Use Before a tool executes Tool arguments, permission decision
Post-Tool Use After a tool executes Tool result, additional context
User Prompt Submitted When user sends a message Nothing (observation only)
Session Start When session begins Nothing (observation only)
Session End When session ends Nothing (observation only)

Quick Start

Register hooks when creating a session:

var hooks = new SessionHooks()
    .setOnPreToolUse((input, invocation) -> {
        System.out.println("Tool: " + input.getToolName());
        return CompletableFuture.completedFuture(
            new PreToolUseHookOutput().setPermissionDecision("allow")
        );
    })
    .setOnPostToolUse((input, invocation) -> {
        System.out.println("Result: " + input.getToolResult());
        return CompletableFuture.completedFuture(null);
    });

var session = client.createSession(
    new SessionConfig()
        .setModel("gpt-4.1")
        .setHooks(hooks)
).get();

Pre-Tool Use Hook

Called before a tool executes. Use this to:

  • Approve, deny, or prompt for tool execution
  • Modify tool arguments
  • Add context for the LLM
  • Suppress tool output from being shown

Input

Field Type Description
getToolName() String Name of the tool being called
getToolArgs() JsonNode Arguments passed to the tool
getCwd() String Current working directory
getTimestamp() long Timestamp in milliseconds

Output

Field Type Description
setPermissionDecision(String) "allow", "deny", "ask" Whether to execute the tool
setPermissionDecisionReason(String) String Reason shown to user/LLM
setModifiedArgs(JsonNode) JsonNode Modified arguments (optional)
setAdditionalContext(String) String Extra context for the LLM
setSuppressOutput(Boolean) Boolean Hide output from display

Example: Security Gate

Block dangerous tool calls:

var hooks = new SessionHooks()
    .setOnPreToolUse((input, invocation) -> {
        String tool = input.getToolName();
        
        // Block file deletion
        if (tool.equals("delete_file")) {
            return CompletableFuture.completedFuture(
                new PreToolUseHookOutput()
                    .setPermissionDecision("deny")
                    .setPermissionDecisionReason("File deletion is not allowed")
            );
        }
        
        // Require confirmation for shell commands
        if (tool.equals("run_terminal_cmd")) {
            return CompletableFuture.completedFuture(
                new PreToolUseHookOutput()
                    .setPermissionDecision("ask")
            );
        }
        
        // Allow everything else
        return CompletableFuture.completedFuture(
            new PreToolUseHookOutput().setPermissionDecision("allow")
        );
    });

Example: Argument Modification

Inject context into tool arguments:

var hooks = new SessionHooks()
    .setOnPreToolUse((input, invocation) -> {
        if (input.getToolName().equals("search_code")) {
            // Add project root to search path
            ObjectMapper mapper = new ObjectMapper();
            ObjectNode modifiedArgs = mapper.createObjectNode();
            modifiedArgs.put("path", "/my/project/src");
            modifiedArgs.set("query", input.getToolArgs().get("query"));
            
            return CompletableFuture.completedFuture(
                new PreToolUseHookOutput()
                    .setPermissionDecision("allow")
                    .setModifiedArgs(modifiedArgs)
            );
        }
        return CompletableFuture.completedFuture(
            new PreToolUseHookOutput().setPermissionDecision("allow")
        );
    });

Post-Tool Use Hook

Called after a tool executes. Use this to:

  • Log tool results
  • Modify the result shown to the LLM
  • Add additional context based on results
  • Suppress output from display

Input

Field Type Description
getToolName() String Name of the tool that was called
getToolArgs() JsonNode Arguments that were passed
getToolResult() String Result from the tool
getCwd() String Current working directory
getTimestamp() long Timestamp in milliseconds

Output

Field Type Description
setModifiedResult(String) String Modified result for the LLM
setAdditionalContext(String) String Extra context for the LLM
setSuppressOutput(Boolean) Boolean Hide output from display

Example: Result Logging

Log all tool executions:

var hooks = new SessionHooks()
    .setOnPostToolUse((input, invocation) -> {
        System.out.printf("[%d] %s completed%n", 
            input.getTimestamp(), 
            input.getToolName());
        System.out.println("Result: " + input.getToolResult());
        
        return CompletableFuture.completedFuture(null);
    });

Example: Result Enrichment

Add context to file read results:

var hooks = new SessionHooks()
    .setOnPostToolUse((input, invocation) -> {
        if (input.getToolName().equals("read_file")) {
            String context = "Note: This file was last modified 2 hours ago.";
            return CompletableFuture.completedFuture(
                new PostToolUseHookOutput()
                    .setAdditionalContext(context)
            );
        }
        return CompletableFuture.completedFuture(null);
    });

User Prompt Submitted Hook

Called when the user submits a prompt, before the LLM processes it. This is an observation hook - you cannot modify the prompt.

Input

Field Type Description
getPrompt() String The user's prompt text
getTimestamp() long Timestamp in milliseconds

Output

Return null - this hook is observation-only.

Example: Prompt Logging

var hooks = new SessionHooks()
    .setOnUserPromptSubmitted((input, invocation) -> {
        System.out.println("User asked: " + input.getPrompt());
        
        // Track prompts for analytics
        analytics.track("user_prompt", Map.of(
            "sessionId", invocation.getSessionId(),
            "promptLength", input.getPrompt().length()
        ));
        
        return CompletableFuture.completedFuture(null);
    });

Session Start Hook

Called when a session starts (either new or resumed).

Input

Field Type Description
getSource() String "new" or "resumed"
getTimestamp() long Timestamp in milliseconds

Output

Return null - this hook is observation-only.

Example: Session Initialization

var hooks = new SessionHooks()
    .setOnSessionStart((input, invocation) -> {
        System.out.println("Session started: " + invocation.getSessionId());
        System.out.println("Source: " + input.getSource());
        
        // Initialize session-specific resources
        sessionResources.put(invocation.getSessionId(), new ResourceManager());
        
        return CompletableFuture.completedFuture(null);
    });

Session End Hook

Called when a session ends.

Input

Field Type Description
getReason() String Why the session ended
getTimestamp() long Timestamp in milliseconds

Output

Return null - this hook is observation-only.

Example: Session Cleanup

var hooks = new SessionHooks()
    .setOnSessionEnd((input, invocation) -> {
        System.out.println("Session ended: " + input.getReason());
        
        // Clean up session resources
        ResourceManager resources = sessionResources.remove(invocation.getSessionId());
        if (resources != null) {
            resources.close();
        }
        
        return CompletableFuture.completedFuture(null);
    });

Complete Example

Combining multiple hooks for comprehensive session control:

import com.github.copilot.sdk.*;
import com.github.copilot.sdk.json.*;
import java.util.concurrent.CompletableFuture;

public class HooksExample {
    public static void main(String[] args) throws Exception {
        try (var client = new CopilotClient()) {
            client.start().get();
            
            var hooks = new SessionHooks()
                // Security: control tool execution
                .setOnPreToolUse((input, invocation) -> {
                    System.out.println("→ " + input.getToolName());
                    
                    // Deny dangerous operations
                    if (input.getToolName().contains("delete")) {
                        return CompletableFuture.completedFuture(
                            new PreToolUseHookOutput()
                                .setPermissionDecision("deny")
                                .setPermissionDecisionReason("Deletion not allowed")
                        );
                    }
                    
                    return CompletableFuture.completedFuture(
                        new PreToolUseHookOutput().setPermissionDecision("allow")
                    );
                })
                
                // Logging: track tool results
                .setOnPostToolUse((input, invocation) -> {
                    System.out.println("← " + input.getToolName() + " completed");
                    return CompletableFuture.completedFuture(null);
                })
                
                // Analytics: track user prompts
                .setOnUserPromptSubmitted((input, invocation) -> {
                    System.out.println("User: " + input.getPrompt());
                    return CompletableFuture.completedFuture(null);
                })
                
                // Lifecycle: initialization and cleanup
                .setOnSessionStart((input, invocation) -> {
                    System.out.println("Session started (" + input.getSource() + ")");
                    return CompletableFuture.completedFuture(null);
                })
                .setOnSessionEnd((input, invocation) -> {
                    System.out.println("Session ended: " + input.getReason());
                    return CompletableFuture.completedFuture(null);
                });
            
            var session = client.createSession(
                new SessionConfig()
                    .setModel("gpt-4.1")
                    .setHooks(hooks)
            ).get();
            
            var response = session.sendAndWait("List files in /tmp").get();
            System.out.println(response.getData().getContent());
            
            session.close();
        }
    }
}

Hook Invocation Object

All hook handlers receive a HookInvocation object as the second parameter:

Method Description
getSessionId() The session ID where the hook was triggered

This allows you to correlate hooks with specific sessions when managing multiple concurrent sessions.


Error Handling

If a hook throws an exception, the SDK logs the error and continues with default behavior:

  • Pre-tool hooks default to allowing execution
  • Post-tool hooks have no effect on the result
  • Lifecycle hooks are observation-only

To handle errors gracefully in your hooks:

.setOnPreToolUse((input, invocation) -> {
    try {
        // Your logic here
        return CompletableFuture.completedFuture(
            new PreToolUseHookOutput().setPermissionDecision("allow")
        );
    } catch (Exception e) {
        logger.error("Hook error", e);
        // Fail-safe: deny if something goes wrong
        return CompletableFuture.completedFuture(
            new PreToolUseHookOutput()
                .setPermissionDecision("deny")
                .setPermissionDecisionReason("Internal error")
        );
    }
})

See Also