Claude Code Hooks Configuration

When to Use This Skill

Use this skill when...	Use plugin-settings instead when...
Wiring a SessionStart, PreToolUse, PostToolUse, or Stop hook into .claude/settings.json	Storing per-project plugin state in .claude/plugin-name.local.md
Setting explicit hook timeouts to prevent "Hook cancelled" errors	Configuring user-tunable plugin behaviour exposed by skill authors
Choosing between command, http, prompt, and agent hook types	Persisting agent state between sessions outside the hook lifecycle
Debugging a hook that fires inconsistently or is silently skipped	Configuring per-project rather than per-session behaviour

Core Expertise

Configure Claude Code lifecycle hooks (all 17 events including SessionStart, Stop, PreToolUse, PostToolUse, PostToolUseFailure, PermissionRequest, WorktreeCreate, TeammateIdle, TaskCompleted, ConfigChange, and more) with proper timeout settings to prevent "Hook cancelled" errors during session management.

Hook Events

Hook	Trigger	Default Timeout
SessionStart	When Claude Code session begins	600s (command)
SessionEnd	When session ends or /clear runs	600s (command)
Stop	When main agent stops responding	600s (command)
SubagentStop	When a subagent (Task tool) finishes	600s (command)
PreToolUse	Before a tool executes	600s (command)
PostToolUse	After a tool completes	600s (command)
PostToolUseFailure	After a tool execution fails	600s (command)
PermissionRequest	When Claude requests permission for a tool	600s (command)
WorktreeCreate	New git worktree created via EnterWorktree	600s (command)
WorktreeRemove	Worktree removed after session exits	600s (command)
TeammateIdle	Teammate in agent team goes idle	600s (command)
TaskCompleted	Task in shared task list marked complete	600s (command)
ConfigChange	Claude Code settings change at runtime	600s (command)

Default timeouts: command = 600s, prompt = 30s, agent = 60s. Always set explicit timeouts — it documents intent.

Hook Types

Type	How It Works	Default Timeout	Use When
command	Runs a shell command, reads stdin, returns exit code/JSON	600s	Deterministic rules
http	Sends hook data to an HTTPS endpoint	30s	Remote/centralized policy
prompt	Single-turn LLM call, returns {ok: true/false}	30s	Judgment on hook input data
agent	Multi-turn subagent with tool access, returns {ok: true/false}	60s	Verification needing file/tool access

Async and Once

• async: true on command hooks: fire-and-forget, does not block the operation
• once: true on any hook handler: runs only once per session, subsequent triggers skipped

For full event reference, schemas, and examples, see .claude/rules/hooks-reference.md.

Common Issue: Hook Cancelled Error

SessionEnd hook [bash ~/.claude/session-logger.sh] failed: Hook cancelled

Root cause: Hook execution exceeds the configured timeout. With the 2.1.50 default of 10 minutes, this is now less common — but explicitly setting "timeout" in your hook config is still recommended.

Solutions (in order of preference):

1. Background subshell - Run slow operations in background, exit immediately
2. Explicit timeout - Add timeout field to hook configuration

Hook Configuration

Location

Hooks are configured in .claude/settings.json:

• User-level: ~/.claude/settings.json
• Project-level: <project>/.claude/settings.json

Structure with Timeout

{
  "hooks": {
    "SessionEnd": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "command",
            "command": "~/.claude/session-logger.sh",
            "timeout": 120
          }
        ]
      }
    ],
    "SessionStart": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "command",
            "command": "~/.claude/session-setup.sh",
            "timeout": 180
          }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "command",
            "command": "~/.claude/stop-hook-git-check.sh",
            "timeout": 30
          }
        ]
      }
    ]
  }
}

Timeout Guidelines

Hook Type	Recommended Timeout	Use Case
SessionStart	120–300s	Tests, linters, dependency checks
SessionEnd	60–120s	Logging, cleanup, state saving
Stop / SubagentStop	30–60s	Git status checks, quick validations
PreToolUse	10–30s	Quick validations
PostToolUse	30–120s	Logging, notifications
PermissionRequest	5–15s	Keep fast for good UX

Fixing Timeout Issues

Recommended: Background Subshell Pattern

The most portable and robust solution is to run slow operations in a background subshell and exit immediately:

#!/bin/bash
# ~/.claude/session-logger.sh
# Exits instantly, work continues in background

(
  # All slow operations go here
  echo "$(date): Session ended" >> ~/.claude/session.log
  curl -s -X POST "https://api.example.com/log" -d "session_end=$(date)"
  # Any other slow work...
) &>/dev/null &

exit 0

Why this works:

• ( ) creates a subshell for the commands
• & runs the subshell in background
• &>/dev/null prevents stdout/stderr from blocking
• exit 0 returns success immediately

Comparison of approaches:

Approach	Portability	Speed	Notes
( ) &	bash, zsh, sh	Instant	Recommended
disown	Bash-only	Instant	Not POSIX
nohup	POSIX	Slight overhead	Overkill for hooks

Alternative: Increase Timeout

If you need synchronous execution, add explicit timeout to settings:

cat ~/.claude/settings.json | jq '.hooks'
# Edit to add "timeout": <seconds> to each hook

Script Optimization Patterns

Optimization	Pattern
Background subshell	( commands ) &>/dev/null &
Fast test modes	--bail=1, -x, --dots
Skip heavy operations	Conditional execution
Parallel execution	Use & and wait

Related: Starship Timeout

If you see:

[WARN] - (starship::utils): Executing command "...node" timed out.

This is a separate starship issue. Fix by adding to ~/.config/starship.toml:

command_timeout = 1000  # 1 second (default is 500ms)

For slow node version detection:

[nodejs]
disabled = false
detect_files = ["package.json"]  # Skip .nvmrc to speed up detection

[command]
command_timeout = 2000  # Increase if still timing out

Agentic Optimizations

Context	Command
View hooks config	cat ~/.claude/settings.json | jq '.hooks'
Test hook script	time bash ~/.claude/session-logger.sh
Find slow operations	bash -x ~/.claude/session-logger.sh 2>&1 | head -50
Check starship config	starship config

Quick Reference

Setting	Location	Default
Hook timeout	.claude/settings.json → hook → timeout	10 minutes (600s) since 2.1.50
Starship timeout	~/.config/starship.toml → command_timeout	500ms
Node detection	~/.config/starship.toml → [nodejs]	Auto

Error Handling

Error	Cause	Fix
Hook cancelled	Timeout exceeded	Add explicit "timeout" (e.g. "timeout": 120)
Hook failed	Script error	Check exit code, add error handling
Command not found	Missing script	Verify script path and permissions
Permission denied	Script not executable	chmod +x ~/.claude/script.sh

Best Practices

1. Use background subshell - Wrap slow operations in ( ) &>/dev/null & and exit 0
2. Set explicit timeouts - Add timeout field for hooks requiring synchronous execution
3. Test hook timing - Use time bash ~/.claude/script.sh to measure execution
4. Redirect all output - Use &>/dev/null to prevent blocking on stdout/stderr
5. Apply /hooks menu - Use Claude Code's hook menu to reload settings after changes