35 hacks, newest first
Ask Claude to log architecture decisions and TODOs into a repo file that can be re-read after /compact.
Create a skill that forces a red→green verification loop so Claude cannot claim done without running the suite.
When Claude repeats a project gotcha, end your prompt with an instruction to write a CLAUDE.md rule so it never happens again.
Inspect `/usage` (alias `/cost`) for per-skill, subagent, and MCP breakdowns, then cut the expensive outliers.
Spin up a subagent for codebase search or verification so only the conclusion returns to your main context.
Proactively run `/compact` with focus instructions so summaries keep the work that matters instead of stale debugging.
Schedule an early Claude Code run so your usage window resets around lunch instead of mid-afternoon.
Write the plan in one session, then open a fresh Claude with no context and ask it to review the plan like a staff engineer.
Turn on `/sandbox` so OS-level filesystem and network boundaries replace many per-command Bash prompts.
Use /init to generate a starter CLAUDE.md, then delete anything inaccurate or obvious before committing.
Double-Esc /rewind back to after useful file reads, then re-prompt with what you learned—without the failed approach still in context.
Instruct Claude (and yourself) to search with rg/Glob first so giant file reads do not blow the context window.
Run prettier/ruff/gofmt automatically after Edit/Write tools so Claude never leaves unformatted files.
Cycle into plan mode (Shift+Tab) so Claude explores and proposes an approach for approval before editing.
Put domain-specific instructions in path-scoped `.claude/rules/` so they load only when Claude touches matching files.
Build slash skills that accept issue numbers, branch names, or formats via $ARGUMENTS / $0 so one command covers many cases.
Use `claude --worktree` (or git worktrees) so multiple agents edit isolated checkouts without colliding.
Shift specialized playbooks into on-demand skills so every session does not pay for instructions you rarely need.
Trim project memory to under ~200 lines so Claude follows critical rules instead of drowning them in noise.
Define custom subagents with `isolation: worktree` so delegated edits happen on a temporary branch checkout.
Run /context to break down token consumers—CLAUDE.md, MCP tools, conversation—and prune the expensive ones.
Add a deterministic PreToolUse hook that rejects destructive shell patterns before Claude can run them.
Use `claude --print --output-format json` in GitHub Actions to automate reviews or summaries without an interactive TTY.
Add the GitHub MCP server so Claude can cross-reference issues, PRs, and repo state without leaving the session.