菜单
Analyzes current conversation context and identifies what can be safely dropped. Triggers a compressed summary that replaces bloated history. Invoke with /trim. Average input token savings: 55–70% on long sessions.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Creates a structured session checkpoint that captures all critical context in minimal tokens. Use /checkpoint to save state before starting a new sub-task, switching files, or when context window is filling up. Typical compression: replaces 8,000–40,000 input tokens with ~400–800 tokens.
Output mode that shows only changed lines as diffs instead of rewriting full files. Cuts output tokens 60–80% on code edits. Works with Claude Code, Codex, Cursor, Windsurf. Activate with /diff-only. Deactivate with /diff-off or "show full file".
Read files surgically — grep first, read only relevant sections. Prevents loading entire large files when only a few lines are needed. Activates automatically when smart-read mode is on. Input token savings: 40–90% depending on file size and task scope.
| name | context-trim |
| description | Analyzes current conversation context and identifies what can be safely dropped. Triggers a compressed summary that replaces bloated history. Invoke with /trim. Average input token savings: 55–70% on long sessions. |
When invoked with /trim, perform the following:
Audit the current conversation history silently.
Identify what is still needed:
Discard (do not include in summary):
Output a compressed context block in this format:
=== CONTEXT CHECKPOINT ===
Goal: [one sentence]
Files in scope: [list]
Decisions locked:
- [decision 1]
- [decision 2]
Open issues:
- [issue 1]
Next step: [what we were about to do]
=== END CHECKPOINT ===
Context trimmed. Saved ~[X]% of prior context. Continuing from checkpoint.
If conversation exceeds ~30 turns or context feels bloated, suggest:
Context is getting long. Run /trim to compress and save ~[estimate]% input tokens.