Hooks / Automations

What it means

The product listens for events (file change, webhook, cron) and starts an agent or skill with fixed context, guardrails, and notification on completion.

Why designers should care

Automation UX needs trigger transparency (“Runs on every push to main”), output routing (Slack vs email), failure alerts, and off switches—silent hooks erode trust.

Example

A Cursor hook runs an accessibility skill on every PR diff and posts a comment with blocking vs advisory findings; designers tune the skill monthly.

Common mistakes

• • Hooks with no audit log of what ran and what changed.
• • Automations that fire on every keystroke, burning tokens and annoying users.
• • No dry-run or preview before enabling a hook on a shared repo.

Related terms

• Agent Skill: A skill defines when to activate, what context to gather, how to execute, and what “done” looks like; the agent reads it at runtime (often from a SKILL.md file).
• Workflow: Defined stages (inputs → generate → review → export) often spanning people, systems, and multiple model calls with clear deliverables.
• Project Rules: Rules encode non-negotiables: coding standards, design tokens, refusal policies, file paths to avoid, and review requirements loaded automatically at task start.
• Action Receipt: Post-action UI: “Archived 14 contacts on Jun 22 at 2:14 PM” with links to affected objects and an undo or support path.
• Loop Engineering: A loop is plan → act → evaluate → revise. Engineering the loop means defining max iterations, success criteria, checkpoints, and what gets logged each pass.