本文由 AI 分析生成
建立時間： 2026-05-30 來源： https://x.com/regent0x_/status/2057419591618302029
Summary
regent0x_ describes going from 23 plugins/8 skills/5 MCP servers to minimal setup after discovering 62,000 tokens were consumed before typing a single prompt. The article catalogs built-in Claude Code commands that replace common plugins: /context, /compact, /clear (context management), /resume, /rewind, Esc+Esc (session control), /cost, /model, /effort (cost/model control), /init, .claudeignore (project setup), /review, /plan (code quality), custom .claude/commands/ (replace most plugins).
regent0x_ 描述從 23 個插件縮減到最小配置後，session 長度延長 3 倍、token 費用減半。文章整理了替代插件的內建指令，核心原則：每個插件都在 context window 中添加基礎負擔，即使閒置也不例外。建議每次 session 開始前執行 /context，基礎佔用超過 30k tokens 就代表有結構性問題。
Key Points
- 62k tokens consumed before first prompt = 31% of 200k context gone before any work
- /context: shows full token breakdown by category
- /compact: compresses conversation history with surgical focus options — most underused command
- Esc+Esc: scrollable checkpoint list for full undo — hidden superpower
- /model: switch Opus/Sonnet mid-session (Sonnet for 80% of work)
- .claudeignore: keeps irrelevant files out of context (replaces file-reading optimization plugins)
- .claude/commands/*.md: custom workflows with zero baseline token overhead
Insights
The math is simple but rarely stated: every plugin adds baseline overhead whether used or not. 62k → 6k baseline tokens = 156k tokens available for work instead of 138k. The practical rule (run /context first, >30k baseline = structural problem) is actionable. The key counterintuitive insight is that most “plugin functionality” is just markdown files with instructions — building it as a .claude/commands/ file costs nothing at baseline while doing the same thing.
Connections
Raw Excerpt
“The math is simple: fewer plugins = more context = longer sessions = better output. Sessions that used to die at 30 minutes now run for 3+ hours.”