Hooks

• Hooks are user-defined handlers that run at deterministic points in an agent’s lifecycle
• Use cases
  • Observability
  • Guardrails
  • Automation like formatting/linting/type-checks/testing

Location of Hooks

• Claude Code: .claude/settings.json
• GitHub Copilot: .github/hooks/NAME.json

Hooks Structure

• Not Universal
• Copilot: https://docs.github.com/en/copilot/how-tos/copilot-cli/customize-copilot/use-hooks
• General Events
  • Session Start
  • User Prompt Submit
  • Pre-Tool Use
  • Post-Tool Use
  • Stop
  • Session End
  • Error Occurred
• Use cases
  • Log Tool call: Pre-Tool Use
  • Run Linter after edit: Post-Tool Use
  • Block dangerous commands: Pre-Tool Use
  • Inject repo context: User Prompt Submit
  • Persist conversation state: Stop/Session End

Example: Hook to Log Tool Input/Output

• .github/hooks

{
  "version": 1,
  "hooks": {
    "preToolUse": [
      {
        "type": "command",
        "bash": "./scripts/pre-tool-log-raw.sh",
        "powershell": "./scripts/pre-tool-log-raw.ps1",
        "cwd": ".github/hooks",
        "timeoutSec": 10
      }
    ],
    "postToolUse": [
      {
        "type": "command",
        "bash": "./scripts/post-tool-log-raw.sh",
        "powershell": "./scripts/post-tool-log-raw.ps1",
        "cwd": ".github/hooks",
        "timeoutSec": 10
      }
    ]
  }
}

• pre-tool-log-raw.sh

#!/bin/bash
set -euo pipefail
 
INPUT="$(cat)"
SCRIPT_DIR="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")" && pwd)"
HOOK_ROOT="$(cd -- "$SCRIPT_DIR/.." && pwd)"
LOG_DIR="$HOOK_ROOT/logs"
RAW_LOG_FILE="$LOG_DIR/pre-tool-use-raw.jsonl"
 
mkdir -p "$LOG_DIR"
chmod 700 "$LOG_DIR"
 
# Log raw hook payload exactly as received on stdin.
printf '%s\n' "$INPUT" >> "$RAW_LOG_FILE"
 
exit 0

• post-tool-log-raw.sh

#!/bin/bash
set -euo pipefail
 
INPUT="$(cat)"
SCRIPT_DIR="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")" && pwd)"
HOOK_ROOT="$(cd -- "$SCRIPT_DIR/.." && pwd)"
LOG_DIR="$HOOK_ROOT/logs"
RAW_LOG_FILE="$LOG_DIR/post-tool-use-raw.jsonl"
 
mkdir -p "$LOG_DIR"
chmod 700 "$LOG_DIR"
 
# Log raw hook payload exactly as received on stdin.
printf '%s\n' "$INPUT" >> "$RAW_LOG_FILE"
 
exit 0

Windows

• pre-tool-log-raw.ps1

$ErrorActionPreference = "Stop"
 
$inputRaw = [Console]::In.ReadToEnd()
$hookRoot = Split-Path -Parent $PSScriptRoot
$logDir = Join-Path $hookRoot "logs"
$rawLogFile = Join-Path $logDir "pre-tool-use-raw.jsonl"
 
if (-not (Test-Path $logDir)) {
  New-Item -ItemType Directory -Path $logDir -Force | Out-Null
}
 
# Log raw hook payload exactly as received on stdin.
Add-Content -Path $rawLogFile -Value $inputRaw
 
exit 0

• post-tool-log-raw.ps1

$ErrorActionPreference = "Stop"
 
$inputRaw = [Console]::In.ReadToEnd()
$hookRoot = Split-Path -Parent $PSScriptRoot
$logDir = Join-Path $hookRoot "logs"
$rawLogFile = Join-Path $logDir "post-tool-use-raw.jsonl"
 
if (-not (Test-Path $logDir)) {
  New-Item -ItemType Directory -Path $logDir -Force | Out-Null
}
 
# Log raw hook payload exactly as received on stdin.
Add-Content -Path $rawLogFile -Value $inputRaw
 
exit 0

Observations on Logs

• All MCP related tools have naming syntax mcp_${server_name}_${tool_name}
  • For example
    • mcp_playwright_browser_tabs
    • mcp_playwright_browser_navigate
    • mcp_playwright_browser_click
  • For all Playwright MCP tools refer: MCP_Playwright
• These are internal tools which Github Copilot (VSCode) has
  • todo
    • Manage and track items for Task planning
  • browser
    • Open and interact with integrated browser pages
  • web
    • Fetch information from website or GitHub
  • edit (Edit files)
  • read (Read files)
  • search (Search files)
  • execute (Execute code and applications)
  • vscode
    • Use VSCode features/extensions/API
    • Ask structured questions to clarify on agent run
  • agent
    • Delegate Tasks to other agents
• Example internal tool names from logs
  • run_in_terminal
  • navigate_page
  • open_browser_page
  • type_in_page
  • click_element
  • read_page
  • read_file