agent-starter: enforcement hooks for Claude Code that catch common AI agent mistakes at write time

Short summary

agent-starter is a Claude Code plugin that installs enforcement hooks to prevent common AI coding agent failures: dangerous git commands, silent error swallowing, hallucinated imports, and dropped awaits. It runs lint on every file write so agents fix mistakes while context is still warm, rather than waiting for CI. The post details each hook's implementation and rationale.

• •Plugin installs hooks that block destructive git commands and silent error swallowing in agent-written code
• •Lint-on-edit hook catches hallucinated imports and async issues at write time, not in CI
• •Exemptions like // silent-ok live in the diff so they show up in code review

Generated with AI, which can make mistakes.