Research

Delegate noisy investigation work to subagents so your context window stays clean. When you need to answer "how does X work" or "what broke in this test run" by reading twenty files, send a subagent to do the digging and get back a distilled answer with evidence instead of polluting your context with seventeen irrelevant files. The key insight is simple: your context is your most limited resource, and reading everything to find the relevant bits permanently degrades your subsequent decisions. Use this liberally for log analysis, codebase surveys, and PR summaries. Skip it when you need to edit the files anyway or when it's just a quick grep.

Install to Claude Code

npx -y skills add warpdotdev/common-skills --skill research --agent claude-code