{"id":"fcca1f04-b9fa-4af1-9165-f551d1d535a2","shortId":"PjA76R","kind":"skill","title":"智能上下文压缩器","tagline":"采用 Rolling Summary 策略压缩对话历史，在清空历史前保留关键信息（任务目标、阶段性结论、任务进展）。","description":"# Execution Instructions\n\n## 核心策略：Rolling Summary（滚动摘要）\n\n**与传统 compactor 的区别**：\n- ❌ 传统方式：直接清空所有历史 → 信息全部丢失\n- ✅ Rolling Summary：先总结关键信息 → 清空历史 → 摘要作为新起点\n\n## 执行步骤\n\n### 原子化压缩操作（Atomic Compression）\n\n现在，压缩操作必须在一个步骤中完成，不再分步进行。\n\n**执行逻辑**：\n1. **Thought (思考)**: 回顾当前对话，在内心生成一份结构化的摘要（包含任务目标、进展、关键数据等）。\n2. **Action (行动)**: 直接调用 `smart_compact(summary=\"...\")`，将生成的摘要作为参数传入。\n\n**摘要模板（供参考）**：\n```text\n## 任务目标\n[用户原始请求]\n\n## 进展\n- [已完成步骤]\n\n## 关键数据\n- [文件路径/配置值]\n\n## 待办\n- [下一步计划]\n```\n\n### 注入摘要\n\n`smart_compact` 执行成功后，会返回包含摘要的确认信息。请将此信息作为新对话的起点。\n\n## 可用工具\n\n### smart_compact\n智能压缩：接收摘要，清空历史，卸载工具。\n**参数**: `summary` (必填) - 当前对话的完整摘要。\n\n### get_compression_status\n获取当前状态和压缩建议。\n\n## 主动调用时机\n\nAgent 应在以下情况**主动**建议调用 compactor：\n\n| 触发条件           | 说明                       |\n| ------------------ | -------------------------- |\n| 对话轮次 > 100 轮  | 上下文过长，影响性能       |\n| 已加载工具 > 50 个 | 工具列表膨胀               |\n| 任务阶段切换       | 完成一个大任务，开始新任务 |\n| 用户明确要求       | \"清理历史\" / \"重置状态\"    |\n\n## 示例流程\n\n```\n用户: \"继续之前的代码分析任务\"\n\nAgent 思考: \"对话已经很长了，我应该压缩一下\"\n\nStep 1 - 生成摘要:\n\"## 任务目标\n分析项目中的内存泄漏问题\n\n## 已完成\n- 使用 codebase_search 定位了 3 个可疑文件\n- 使用 bash 检查了进程内存\n\n## 关键发现\n- memory_pool.py 第 45 行存在未释放的资源\n- 问题出现在 process_data() 函数中\n\n## 待处理\n- 需要修复 memory_pool.py 的问题\"\n\nStep 2 - 执行压缩:\nAction: smart_compact()\nResult: [OK] 已清空历史，已卸载 5 个工具\n\nStep 3 - 继续任务:\n\"根据之前的分析，memory_pool.py 第 45 行存在问题，\n现在让我加载 codebase_search 继续处理...\"\n```\n\n## 重要提醒\n\n1. **压缩前必须生成摘要** - 否则关键信息会丢失\n2. **摘要要简洁但完整** - 保留任务目标、结论、进展\n3. **压缩后主动告知用户** - 让用户知道之前做了什么","tags":["compactor","google","adk","agent","valkryhx","agent-development","agent-skills","agent-swarm","agent-team","agentic-ai","dynamic-skills","google-adk"],"capabilities":["skill","source-valkryhx","skill-compactor","topic-agent-development","topic-agent-skills","topic-agent-swarm","topic-agent-team","topic-agentic-ai","topic-dynamic-skills","topic-google-adk","topic-vibe-coding"],"categories":["google_adk_agent"],"synonyms":[],"warnings":[],"endpointUrl":"https://skills.sh/valkryhx/google_adk_agent/compactor","protocol":"skill","transport":"skills-sh","auth":{"type":"none","details":{"cli":"npx skills add valkryhx/google_adk_agent","source_repo":"https://github.com/valkryhx/google_adk_agent","install_from":"skills.sh"}},"qualityScore":"0.453","qualityRationale":"deterministic score 0.45 from registry signals: · indexed on github topic:agent-skills · 6 github stars · SKILL.md body (1,461 chars)","verified":false,"liveness":"unknown","lastLivenessCheck":null,"agentReviews":{"count":0,"score_avg":null,"cost_usd_avg":null,"success_rate":null,"latency_p50_ms":null,"narrative_summary":null,"summary_updated_at":null},"enrichmentModel":"deterministic:skill-github:v1","enrichmentVersion":1,"enrichedAt":"2026-05-18T19:14:45.140Z","embedding":null,"createdAt":"2026-05-18T13:22:13.273Z","updatedAt":"2026-05-18T19:14:45.140Z","lastSeenAt":"2026-05-18T19:14:45.140Z","tsv":"'1':35,115,167 '100':93 '2':43,143,170 '3':124,155,175 '45':132,160 '5':152 '50':98 'action':44,145 'agent':85,110 'atom':29 'bash':127 'codebas':121,163 'compact':48,65,71,147 'compactor':17,89 'compress':30,81 'data':136 'execut':10 'get':80 'instruct':11 'memory_pool.py':130,140,158 'ok':149 'process':135 'result':148 'roll':3,13,22 'search':122,164 'skill' 'skill-compactor' 'smart':47,64,70,146 'source-valkryhx' 'status':82 'step':114,142,154 'summari':4,14,23,49,77 'text':53 'thought':36 'topic-agent-development' 'topic-agent-skills' 'topic-agent-swarm' 'topic-agent-team' 'topic-agentic-ai' 'topic-dynamic-skills' 'topic-google-adk' 'topic-vibe-coding' '上下文过长':95 '下一步计划':62 '不再分步进行':33 '与传统':16 '个':99 '个可疑文件':125 '个工具':153 '主动':87 '主动调用时机':84 '任务目标':7,54,117 '任务进展':9 '任务阶段切换':101 '会返回包含摘要的确认信息':67 '传统方式':19 '使用':120,126 '供参考':52 '保留任务目标':172 '信息全部丢失':21 '先总结关键信息':24 '关键发现':129 '关键数据':58 '关键数据等':42 '函数中':137 '分析项目中的内存泄漏问题':118 '包含任务目标':40 '卸载工具':75 '压缩前必须生成摘要':168 '压缩后主动告知用户':176 '压缩操作必须在一个步骤中完成':32 '原子化压缩操作':28 '参数':76 '可用工具':69 '否则关键信息会丢失':169 '回顾当前对话':38 '在内心生成一份结构化的摘要':39 '在清空历史前保留关键信息':6 '完成一个大任务':102 '定位了':123 '对话已经很长了':112 '对话轮次':92 '将生成的摘要作为参数传入':50 '工具列表膨胀':100 '已加载工具':97 '已卸载':151 '已完成':119 '已完成步骤':57 '已清空历史':150 '应在以下情况':86 '建议调用':88 '开始新任务':103 '当前对话的完整摘要':79 '影响性能':96 '待办':61 '待处理':138 '必填':78 '思考':37,111 '我应该压缩一下':113 '执行压缩':144 '执行成功后':66 '执行步骤':27 '执行逻辑':34 '接收摘要':73 '摘要作为新起点':26 '摘要模板':51 '摘要要简洁但完整':171 '文件路径':59 '智能上下文压缩器':1 '智能压缩':72 '核心策略':12 '根据之前的分析':157 '检查了进程内存':128 '注入摘要':63 '清理历史':105 '清空历史':25,74 '滚动摘要':15 '现在':31 '现在让我加载':162 '生成摘要':116 '用户':108 '用户原始请求':55 '用户明确要求':104 '的区别':18 '的问题':141 '直接清空所有历史':20 '直接调用':46 '示例流程':107 '第':131,159 '策略压缩对话历史':5 '结论':173 '继续之前的代码分析任务':109 '继续任务':156 '继续处理':165 '获取当前状态和压缩建议':83 '行动':45 '行存在未释放的资源':133 '行存在问题':161 '触发条件':90 '让用户知道之前做了什么':177 '说明':91 '请将此信息作为新对话的起点':68 '轮':94 '进展':41,56,174 '配置值':60 '采用':2 '重置状态':106 '重要提醒':166 '问题出现在':134 '阶段性结论':8 '需要修复':139","prices":[{"id":"468db79f-d997-4131-bf50-467adbb00fe0","listingId":"fcca1f04-b9fa-4af1-9165-f551d1d535a2","amountUsd":"0","unit":"free","nativeCurrency":null,"nativeAmount":null,"chain":null,"payTo":null,"paymentMethod":"skill-free","isPrimary":true,"details":{"org":"valkryhx","category":"google_adk_agent","install_from":"skills.sh"},"createdAt":"2026-05-18T13:22:13.273Z"}],"sources":[{"listingId":"fcca1f04-b9fa-4af1-9165-f551d1d535a2","source":"github","sourceId":"valkryhx/google_adk_agent/compactor","sourceUrl":"https://github.com/valkryhx/google_adk_agent/tree/main/skills/compactor","isPrimary":false,"firstSeenAt":"2026-05-18T13:22:13.273Z","lastSeenAt":"2026-05-18T19:14:45.140Z"}],"details":{"listingId":"fcca1f04-b9fa-4af1-9165-f551d1d535a2","quickStartSnippet":null,"exampleRequest":null,"exampleResponse":null,"schema":null,"openapiUrl":null,"agentsTxtUrl":null,"citations":[],"useCases":[],"bestFor":[],"notFor":[],"kindDetails":{"org":"valkryhx","slug":"compactor","github":{"repo":"valkryhx/google_adk_agent","stars":6,"topics":["agent-development","agent-skills","agent-swarm","agent-team","agentic-ai","dynamic-skills","google-adk","vibe-coding"],"license":null,"html_url":"https://github.com/valkryhx/google_adk_agent","pushed_at":"2026-04-23T07:34:09Z","description":"a startup but not simple agent demo using  google adk.","skill_md_sha":"ef7d9c88d6c4a9cade195a7e4c1fdd96977e61fe","skill_md_path":"skills/compactor/SKILL.md","default_branch":"main","skill_tree_url":"https://github.com/valkryhx/google_adk_agent/tree/main/skills/compactor"},"layout":"multi","source":"github","category":"google_adk_agent","frontmatter":{"name":"智能上下文压缩器","description":"采用 Rolling Summary 策略压缩对话历史，在清空历史前保留关键信息（任务目标、阶段性结论、任务进展）。"},"skills_sh_url":"https://skills.sh/valkryhx/google_adk_agent/compactor"},"updatedAt":"2026-05-18T19:14:45.140Z"}}