New tricks in the playbook, v1.5 is published with skills and hooks.

Nine skills now sit in the skills/ folder. You don’t call them. Claude picks up on what you’re trying to do and loads the right one on its own.

Say “can you review my changes before I open a PR?” and code-review kicks in. Paste an error and build-fix kicks in. Say you’re wrapping up and checkpoint kicks in.

The full set: code-review test-writer security-check build-fix docs verify checkpoint context-seed refactor. Each has an instruction file, an output template, and a manual inside skills/. Your existing slash commands in agents/ still work. Skills sit alongside them, not on top.

Five scripts that run on their own during your Claude Code session. You don’t trigger them. They just run.

• guard-destructive.sh — blocks rm -rf, force pushes, DROP TABLE, and similar irreversibles before they run.
• lint-on-write.sh — runs your linter immediately after any file write. Claude sees the output in the same turn.
• test-on-write.sh — finds the paired test file for whatever you just changed and runs it automatically. Async, doesn’t block you.
• docs-nudge.sh — catches you at session end if source files changed and no docs pass ran.
• auto-checkpoint.sh — writes a minimal git-state savepoint to .claude/checkpoints/ every time Claude stops.

All five scripts are in hooks/scripts/. The ready-to-use config is in hooks/settings-template.json. Takes about five minutes to set up. hooks/README.md has the steps.

For QA and frontend: Playwright is now in the active MCP list. It lets Claude control a real browser, click through flows, fill forms, take screenshots, and run end-to-end tests. If you’re backend-only, you can skip this one.

Context7 now has a plugin install path too. If you haven’t set it up yet, that’s the easier way to go.

Full reference → docs/13-hooks-and-automation.md and docs/14-skills.md