What Are Hooks in Claude Code?

What does a hook mean in Claude Code?

A hook in Claude Code is a shell command that runs automatically at a specific point in the coding agent's lifecycle, before or after a tool call, when a session starts, or when the agent stops, configured in a settings file rather than embedded in the agent's own instructions. DataCamp's own guide to the feature describes hooks as offering "deterministic control over what the AI agent does at key moments in its workflow," distinguishing them from prompt-based instructions the model may interpret flexibly: a hook executes a defined shell command automatically, outside the model's own judgment about whether to follow it. A PreToolUse hook, for example, can block a risky operation before it runs at all, while a PostToolUse hook can automatically format, lint, or test code right after a change.

The core distinction is enforcement versus recommendation. A written rule can be missed under pressure. A hook runs every time, regardless of context.

Why hooks matter for product managers

Builders Camp's Building with Claude Code bootcamp lists hooks, CI/CD, and quality gates as one of its six core modules, describing the goal directly: implementing "automation triggers, quality checks, and CI/CD-style workflows to ensure your AI pipelines produce reliable, production-ready output." Its certification quiz draws the distinction that matters most for anyone overseeing this kind of workflow: CLAUDE.md contains recommendations the agent should follow, while hooks are automated controls that enforce rules "regardless of what the agent outputs."

For a PM setting standards for an AI-native engineering team, this distinction changes what "we have a policy for that" actually means. A policy written in a document can be forgotten under deadline pressure. A hook enforcing the same policy cannot be skipped that way.

How hooks are used in practice

Building with Claude Code's practical challenge builds a governance layer exercise directly around this idea: participants design one hook, naming exactly what triggers it and what it prevents, as part of a broader system meant to stop a repeat of a real incident, a broken deploy that shipped because nobody had defined what "done" meant for the AI agent working on it. The challenge caps each governance component at a maximum of three bullet points specifically to force a decision about which controls actually matter most, rather than listing every conceivable check.

A concrete example from that same challenge: a hook that automatically runs the test suite and blocks a commit if any test fails, paired with a rule that any change touching payment logic requires a human review before merging, regardless of what the automated tests report.

How Builders Camp teaches hooks

Building with Claude Code, taught by Guilherme Salgueiro, covers hooks as part of its governance layer alongside quality gates and reusable agent skills, treating all three as the mechanisms that keep an AI coding workflow reliable at scale. Builders Camp's Building your AI Operating System bootcamp applies the same automation and quality control discipline to non-coding PM workflows.

See parallel worktrees for the workflow pattern hooks often need to verify across multiple simultaneous workstreams. See the Building with Claude Code bootcamp for the full curriculum and practical challenge.