← 返回
未分类
中文
Strategic Compact
Manual context compaction at logical workflow boundaries rather than arbitrary auto-compaction. Use after completing exploration phases, before switching maj...
在逻辑工作流边界手动压缩上下文,而非随意的自动压缩。在完成探索阶段后、切换主要任务前使用。
未分类
clawhub
v1.0.1 1 版本 100000 Key: 无需
#latest
概述
版本历史 (1)
安全
概述
Strategic Compact Skill Manual /compact at logical workflow boundaries, not arbitrary auto-compaction points.
When to Activate Running long sessions approaching context limits (200K+ tokens) Multi-phase tasks (research → plan → implement → test) Switching between unrelated tasks within the same session After completing a major milestone before starting new work Quick Start Identify logical phase boundaries in your work Save important findings to files before compacting Use /compact with a context message (e.g., "/compact Now implementing auth") Continue work referencing saved files, not conversation history Key Concepts Auto-compaction problem — Triggers at arbitrary points, often mid-taskStrategic compaction — Compact after exploration/planning, before executionPersistence — Project instructions, git state, and files on disk survive; conversation history and intermediate reasoning don'tCheckpoint discipline — Save learnings to files; use git commits to mark progressToken awareness — 150K → consider; 180K → plan; 200K+ → should have already compactedCommon Usage Most frequent patterns:
Research → build (compact after documentation phase) Exploration → execution (compact after CLAUDE.md is written) Debug → next feature (compact after fix is committed) Failed approach → new strategy (compact to clear dead-end reasoning) References references/compaction-guide.md — Decision table, what survives, patterns, token budgeting, optimization strategiesreferences/tool-integration.md — Script documentation, configuration, integration methodsCompaction command — /compact [optional message]Status check — Check session context size before major operationsMemory files — Save to disk (survives compaction)Git commits — Mark progress (persists across compaction)
版本历史
共 1 个版本
v1.0.1
当前
2026-05-03 10:13 安全 安全
安全检测
腾讯云安全 (Sanbu)
安全,无风险
查看报告
🔗 相关推荐
ai-agent
rez0
用于 AI 代理的浏览器自动化 CLI。当用户需要与网站交互(包括浏览页面、填写表单、点击按钮、截图等)时使用。
★ 871
📥 349,697
dev-programming
djc00p
使用分层升级循环修复Rails PR的CI失败。当Rails拉取请求出现CI失败时使用此技能——包括RSpec测试失败、RuboCop违规等。
★ 0
📥 960
ai-agent
root
帮助用户发现和安装智能体技能,当用户询问如「如何做X」、「找X的技能」、「有能做...的吗」等问题时
★ 1,524
📥 581,232