Awesome list of Claude Code tips, tricks, gotchas
A community-sourced collection of Claude Code tips covering skills-based workflows, spec-driven development, git worktrees, and autonomous coding sessions.
Summary
A New Year's community thread collecting practical Claude Code tips. The OP shares a workflow emphasizing custom skills, spec-driven development, and git worktrees for parallel feature work.
Key Tips from OP
- Next prompt prediction as pseudo-memory - Create custom skills for each workflow part; Claude Code predicts task-to-skill relationships via tab completion
- Commit before worktrees - Dirty changes don't clone to worktrees; commit first or sync after
- Spec-driven > plan mode - Create expectations → requirements → technical design → task breakdown to force deeper thinking
- Autonomous mode + worktree + spec - This combination enables 20-30 minute autonomous sessions that complete 90% of work
- Keep context lean - Avoid jamming context with MCPs; use
/contextcommand to monitor, prune skills regularly - Playwright e2e testing - Build regression harness around key user journeys; use visual testing to avoid faked tests
Workflow Pattern
Create a skill workflow to constrain one feature slice level expectation → specifications → tasks breakdown. This will usually last 1-2 auto compact sessions.
The thread advocates for a structured approach: isolate work in git worktrees, drive implementation from specs rather than ad-hoc prompting, and let autonomous mode handle execution.
Related
See also claude-code-best-practices and writing-a-good-claude-md.