CLAUDE.md is the rulebook.
It gives Claude your standards, conventions, and project context before the work starts, so you stop repeating yourself every session./plan makes Claude think before it builds.
You review the logic first, then approve implementation. That alone can save a lot of wasted output.
/agents helps split independent work into parallel streams.
Useful when multiple tasks can move at the same time without blocking each other.
/skills is for repeatable workflows.
If you keep doing the same thing, document it once and reuse it instead of prompting from scratch every time.
/MCP connects Claude to the tools you already use.
GitHub, Linear, Notion, databases, logs. Better context comes from real sources, not memory or manual copy-pasting.
/compact helps when long sessions start getting noisy.
It trims the conversation history and preserves the signal.
autodream is not a typed command.
It lives under /memory and helps refine project context over time by identifying recurring patterns and themes.
/ralph-loop is not native, but it is worth knowing.
It keeps Claude iterating on the same task until completion. Just make sure you set a maximum iteration limit.
The difference is not "using AI."
It is building structure around it.
No comments:
Post a Comment