Hooks System (Cursor Implementation)

Hook Types

• •beforeShellExecution: Before terminal command execution (validation, safety checks)
• •afterShellExecution: After terminal command execution (log analysis, result parsing)
• •afterFileEdit: After file modification (auto-format, linting, code quality checks)
• •stop: When session/generation ends (final verification, auditing)

Current Hooks (in .cursor/hooks.json & .cursor/hooks/*.sh)

beforeShellExecution (Script: before_cmd.sh)

• •Long-running process alert: Detects long-running commands (npm, cargo, build, etc.) and logs a visible info message (Windows compatible, no tmux enforcement).
• •Git push safety: Pauses execution for 3 seconds before git push to allow user interruption (Ctrl+C) for review.
• •Doc blocker: Blocks creation of unnecessary .md/.txt files via shell commands (preserves README/CLAUDE.md).

afterShellExecution (Script: after_cmd.sh)

• •PR creation logger: Detects gh pr create output, logs the new PR URL, and suggests the review command.

afterFileEdit (Script: after_edit.sh)

• •Prettier: Auto-formats JS/TS/JSX/TSX files immediately after Agent edits.
• •TypeScript check: Runs tsc --noEmit on the project root to catch new type errors in the edited file.
• •console.log warning: Scans the specific edited file for console.log and alerts the Agent.

stop (Script: on_stop.sh)

• •console.log audit: Runs git diff to check all modified files in the session for remaining console.log statements before finishing.

Permissions & Safety

In Cursor settings:

• •Yolo Mode: Use with caution. Disable for exploratory or sensitive operations.
• •.cursorrules: Define project-specific permission boundaries here instead of ~/.claude.json.
• •Terminal Allowlist: Configure allowed commands in Cursor settings to reduce manual approvals.

Task Tracking Best Practices (Composer/Agent)

Use a dedicated TODO.md or Agent Plan to:

• •Track progress on multi-step tasks
• •Verify understanding of instructions
• •Enable real-time steering
• •Show granular implementation steps

Agent Todo list reveals:

• •Out of order steps
• •Missing items
• •Extra unnecessary items
• •Wrong granularity
• •Misinterpreted requirements