Hooks System

todo-continuation-enforcer

Ensures tasks are completed. Monitors todo items and prompts for continuation if tasks are left incomplete.

When it runs: During task execution and completion checks

empty-task-response-detector

Detects when responses are empty or incomplete, helping ensure tasks are properly handled.

When it runs: After receiving agent responses

context-window-monitor

Monitors and manages context window usage. Tracks token usage and warns when approaching limits.

When it runs: Continuously during session

compaction-context-injector

Manages context compaction to stay within token limits. Intelligently compacts old context while preserving important information.

When it runs: When context window approaches limits

preemptive-compaction

Preemptive context compaction that runs before hitting limits. Configurable threshold (default 0.85).

When it runs: When context usage reaches threshold

Configuration: experimental.preemptive_compaction_threshold

anthropic-context-window-limit-recovery

Handles Anthropic-specific context window limits and recovery. Automatically manages context when hitting Anthropic API limits.

When it runs: When using Anthropic models and hitting limits

session-recovery

Automatically recovers from errors, thinking block violations, and interruptions. Never lose your progress.

When it runs: After errors or interruptions

session-notification

Notifies about session events like starts, errors, and completions.

When it runs: On session events

auto_resume

Experimental Automatically resumes session after successful recovery. Extracts the last user message and continues.

When it runs: After successful recovery

Configuration: experimental.auto_resume

comment-checker

Validates code comments. Ensures code is properly documented and comments are meaningful.

When it runs: During code generation and review

thinking-block-validator

Validates thinking blocks to ensure they follow proper format and structure.

When it runs: When processing thinking blocks

grep-output-truncator

Truncates large grep outputs to prevent context window overflow. Intelligently truncates while preserving important matches.

When it runs: After grep operations with large outputs

tool-output-truncator

Manages tool output sizes. Truncates outputs from whitelisted tools (Grep, Glob, LSP, AST-grep) to stay within limits.

When it runs: After tool operations with large outputs

Configuration: Can be configured to truncate ALL tool outputs via experimental.truncate_all_tool_outputs

aggressive_truncation

Experimental More aggressive tool output truncation when token limit is exceeded. Falls back to summarize/revert if insufficient.

When it runs: When token limit is exceeded

Configuration: experimental.aggressive_truncation

directory-agents-injector

Injects directory-specific agents based on directory structure. Automatically enables relevant agents for different parts of your codebase.

When it runs: When entering directories with agent configurations

directory-readme-injector

Adds README context automatically. Injects README files from directories to provide better context.

When it runs: When entering directories with README files

rules-injector

Injects custom rules into the agent context. Allows project-specific rules and guidelines.

When it runs: At session start and when rules change

think-mode

Enables thinking mode for agents. Allows agents to use thinking blocks for complex reasoning.

When it runs: When thinking mode is requested

claude-code-hooks

Claude-specific hooks for enhanced Claude Code integration.

When it runs: When using Claude models

ralph-loop

Ralph agent loop management. Handles Ralph agent specific workflows.

When it runs: When using Ralph agent

auto-update-checker

Checks for updates automatically. Notifies you when new versions are available.

When it runs: At startup and periodically

Note: The startup-toast hook is a sub-feature of this hook.

startup-toast

Shows startup notifications. Displays welcome messages and important information at startup.

When it runs: At startup

Note: This is a sub-feature of auto-update-checker. Disable separately to keep update checking but remove toast.

keyword-detector

Detects keywords in conversations. Can trigger specific behaviors based on detected keywords.

When it runs: During conversation processing

agent-usage-reminder

Reminds about agent usage. Provides helpful reminders about available agents and their capabilities.

When it runs: Periodically during sessions

non-interactive-env

Handles non-interactive environments. Adapts behavior for CI/CD and automated environments.

When it runs: When in non-interactive mode

interactive-bash-session

Manages interactive bash sessions. Handles interactive terminal sessions properly.

When it runs: During interactive terminal operations

empty-message-sanitizer

Cleans empty messages. Removes or handles empty messages in conversations.

When it runs: When processing messages