20:"$Sreact.view_transition" 21:I[9754,["/_next/static/chunks/2t36f5p4pjzlb.js","/_next/static/chunks/1gyqeidjm5eoa.js","/_next/static/chunks/0_fjnxc4tt_l_.js","/_next/static/chunks/1ymt1shyhmu-4.js","/_next/static/chunks/3kv2zf0i7s2wc.js","/_next/static/chunks/1vs_3g_g75zmk.js","/_next/static/chunks/2g4dh802s8oz3.js","/_next/static/chunks/1bc3f0shukf1j.js","/_next/static/chunks/192kgs9v_t14j.js","/_next/static/chunks/0z6pw0v5c_lx5.js","/_next/static/chunks/08r60e6icmtkc.js","/_next/static/chunks/3trx6c0i7_x-3.js","/_next/static/chunks/40sro5bzp9bi_.js","/_next/static/chunks/1wgf3tk38y6cf.js"],"ProjectDetail"] 22:T4e93, context-mode · 记忆地层

context-mode

把上下文压实成记忆地层

为 AI 编码代理优化上下文窗口的 MCP 服务器

原始输出不该占满那唯一一段窗口。Context Mode 用沙箱工具把原始数据排除在窗外、把会话磨成岩层，再用检索把真正相关的那一条抽回地表——节省、持久、可归因。

访问 context-mode.com GitHub 仓库

作者 mksglu TypeScript Star 19916 Fork 1432

会话如何被压实成记忆

每一段会话都像岩层一样被写入：砂箱把膨胀的原始输出排除在窗外，SQLite 把余下的事件一层层打磨埋藏。越往下的层越薄、越被压缩。

a沙箱挡在窗外原始工具输出先在沙箱里被处理，只有修剪后的线索进入上下文窗口，压缩占用至 98% 的减少。
b逐层压薄信息量越低的层越窄；压缩黑云把表层截断，旧内容沉到下面，而不是直接抹掉。
c只挖有必要的压薄不等于丢弃——被埋的层仍以事件形式保留，等检索把真正相关的那条抽回。

上层的「当前窗口」最宽、信息最满；下层连续收窄。全页地层剖面从宽到窄，正是上下文被逐层压实、记忆被可持续保存的示意。

FTS5 为事件编号

每一层不是石堆，而是被编号过的索引。Context Mode 用 SQLite + FTS5 为会话事件建立全文索引，让每一条被埋的记忆都保留可检索的地址——压缩之后，没有一层真正丢失。

i事件即岩层每个会话事件被写入 SQLite，作为岩层持久保存。
iiFTS5 编号全文索引让每一层都可通过关键词被定位，而不是一团无法回访的堆积物。
iiiBM25 就绪压实的表层之下，检索的钻头随时可以沿编号下探。

BM25 把相关回抽

压缩从来不是只出不进。真正的关键在于：当对话压缩后需要回忆，一条检索线索顺着索引钻下最深的地层，把相关的那几条抽回地表。

r1BM25 搜索检索请求命中 FTS5 索引下的索引层，按相关性重排序候选项。
r2从底回升签名视觉中那条磷光金线，就是从最深层蜿蜒回到当前窗口的检索路径。
r3只抽必要的回到表层的只是与当下问题相关的事件，而不是整块被埋的旧层。

层与层之间是一条矿脉：检索线悬浮在切面之上，不与岩层混融——记忆被埋，但永不失联。

会话连续性 · 持久记忆

长时间的重组不该清零一切。会话记忆被持久地埋进岩层，跨会话仍能沿索引回抽。矿脉从底回升到顶的方向，正是这条连续性本身的走向——需要的团队可以在长会话之间维持住上下文。

s持久记忆会话事件不是一次性草稿，而是写入地层、可反复回访的记忆。
s连续性即成层对话压缩后再续，检索沿既有索引把旧相关抽回，仿佛没有断过。

17 平台统一路由

1MCP + hooks通过 MCP 与 hooks 强制执行路由，覆盖 17 个平台。
2Claude Code优先标记的代表平台之一；Gemini CLI、Codex 等也在列。
3同一套地层无论落在哪个平台，会话都被压入同一套索引层，检索口径保持一致。

地层不因平台而异：同一套沙箱、同一套 FTS5 索引、同一条回抽矿脉，被 17 个平台共享。

节省与 FinOps 归因

压实的层也该被计量。最新的 v1.0.169 改进了上下文节省报告与 FinOps 成本归因的准确性——每一层被埋下的数据，最终都能换算成可核算的账单。

上下文减少

98%

沙箱把原始输出排除在窗外

覆盖平台

通过 MCP 与 hooks 统一路由

MCP 工具

含沙箱工具与元工具

FinOps

v1.0.169

更准确的节省与成本归因

23:T4df8, context-mode · The Memory Strata

context-mode

Compacting context into memory strata

An MCP server for optimizing AI coding agent context

Raw output should never fill that single window. Context Mode keeps raw data out of the context window with sandbox tools, grinds sessions into strata, and drills the one genuinely relevant vein back to the surface - saving context, persisting memory, and keeping it all attributable.

Visit context-mode.com GitHub repository

By mksglu TypeScript Stars 19916 Forks 1432

How sessions get compacted into memory

Every session is written like a layer of rock: the sandbox keeps bloated raw output out of the window, and SQLite grinds what remains into thinner, deeper beds the further down you go.

aThe sandbox gates the windowRaw tool output is handled inside a sandbox first; only trimmed clues enter the context window, cutting usage by up to 98%.
bCompacted layer by layerThe less informative a layer, the thinner it becomes; compaction shears the top, and old matter sinks instead of vanishing.
cBuried is not deletedCompressed layers persist as events, waiting for retrieval to pull the genuinely relevant band back up.

The current window above is widest and fullest; things narrow continuously toward the bottom. The strata profile across the page tracks exactly how context gets compacted and held.

FTS5 assigns events an index

Each layer is not a stone heap but an indexed deposit. Context Mode uses SQLite with FTS5 to build a full-text index over session events, so every buried memory keeps an address to retrieve - after compaction, no layer is truly lost.

iEvents become strataEach session event is written into SQLite and held as a persistent layer.
iiFTS5 labels each bedFull-text indexing locates any layer by keyword rather than leaving an unaddressable pile.
iiiBM25 at the readyBelow the compacted surface, the retrieval drill can always bore down the index.

BM25 drills relevance back

Compaction is never one-way. When a compacted conversation needs to recall, a single retrieval vein runs down the index to the deepest bed and raises the relevant bands back to the surface.

r1BM25 searchA query hits the indexed beds under FTS5 and reranks candidates by relevance.
r2Back up from the bottomThe phosphor-gold line in the signature visual is exactly that retrieval path curling from the deepest layer to the current window.
r3Raise only what mattersOnly events relevant to the question return to the surface, not the whole buried bed.

Between layers runs a single vein: retrieval floats above the cut rather than mixing into the strata - memory stays buried, but never out of reach.

Session continuity · persistent memory

A long session should not reset to zero. Session memory stays buried in the strata and can be raised again along the index across sessions. The vein trending from the bottom to the top is the shape of that continuity itself - teams can carry context through very long sessions.

sPersistent memorySession events are not one-shot drafts; they are deposited as strata and revisited.
sContinuity is a layerAfter compaction, the next turn raises old relevance along existing index beds, as if nothing was interrupted.

Unified routing across 17 platforms

1MCP + hooksRouting is enforced via MCP and hooks across 17 platforms.
2Claude Code firstA flagship among the covered platforms, alongside Gemini CLI, Codex, and others.
3One set of strataWhatever platform you land on, sessions settle into the same indexed beds with matching retrieval semantics.

The strata do not change by platform: one sandbox, one FTS5 index, one returning vein, shared across 17 platforms.

Savings and FinOps attribution

Compacted beds should be measurable too. The latest v1.0.169 improves context-savings reporting and FinOps cost attribution accuracy - every layer buried ultimately resolves into a bill you can trace.

Context reduction

98%

sandbox keeps raw output out

Platforms

routed via MCP and hooks

MCP tools

sandbox tools plus meta-tools

FinOps

v1.0.169

more accurate savings attribution

a:[["$","$20",null,{"enter":{"secondary-forward":"secondary-forward","secondary-back":"secondary-back-enter","default":"none"},"exit":{"secondary-forward":"secondary-forward","secondary-back":"secondary-back-exit","default":"none"},"update":{"secondary-forward":"secondary-forward","secondary-back":"secondary-back-update","default":"none"},"default":"none","children":["$","$L21","cmt2zh1h1012z6271enammdmq:zh-CN:beautified",{"project":{"id":"cmt2zh1h1012z6271enammdmq","slug":"mksglu-context-mode","name":"context-mode","description":"为 AI 编码代理优化上下文窗口的 MCP 服务器","authorName":"mksglu","authorVerification":{"verified":false,"verifiedAt":null},"category":"development_tools","subcategory":"data_ai_tools","experienceUrl":"https://context-mode.com/","qualityScore":90,"featured":false,"officialSite":null,"views":{"original":{"available":true},"official":{"available":false,"reason":"disabled_by_policy"},"beautified":{"available":true,"reason":null},"defaultView":"beautified"},"beautifiedView":{"revisionId":"cmt2zh1ht013b62711raksnvy","kind":"artifact","artifact":{"schemaVersion":"localized-static-html-v4","kind":"static_html","locales":["zh-CN","en"],"defaultLocale":"zh-CN","documents":{"zh-CN":"$22","en":"$23"},"designDirections":{"schemaVersion":"dodot-design-directions-v1","directions":[{"id":"dir-a-copressor","name":"The Compressor · 双寄存器收缩术","projectMetaphor":"上下文窗口是一根固定长度的寄存器；Context Mode 用沙箱工具把原始数据压缩出寄存器，再用检索把相关的抽回来。整页像一次可见的压缩/解压往返。","signatureVisual":"一根上下文预算带：同一行 token 计量，左侧被两根收敛斜线（挖斗）夹成一个收窄指针——压缩前用文本长度计量，压缩后用符号沙箱化标记计量；压缩区用括弧做减法，展开区用检索箭头做加法，一眼呈现 98% 的收缩方向。","heroGeometry":"非对称水平二分：上部是出口（沙箱把膨胀数据倒出，垂直线条向右下倾斜），下部是入口（检索线索从底部反向抽回，线束向右上收敛），两条斜线在版面中央构成一个有限的腰，隐喻寄存器窄口。","sectionSequence":["The Leak · 爆炸式输出吞噬窗口","The Valve · 沙箱把数据挡在窗外","The Recall · FTS5/BM25 把相关抽回","The Route · 17 平台统一强制路由","The Ledger · 上下文节省与 FinOps 归因"],"surfaceModel":"平整色带截面：每个 section 是一条方向性的寄存器带，无卡片层叠；带与带之间用反向斜切衔接，色幅从不重复到渐深，营造一次连续的收缩—回抽动量。","palette":"以墨色寄存器为底（#0b0e14），主色为磷光青（#22d3a7）表示通过/抽出，次色为琥珀（#ffb35c）表示收缩/积蓄，第三色为冷紫（#9a7bff）用于倒出的弃料；留白占位，不用大块光晕。","typography":"衬线标题（Warning 风格的尖锐加粗）搭配等宽副标题（IBM Plex Mono），数字用等宽对齐以表现计量；正文用系统无衬线，行距紧凑，强调刻度感。","desktopComposition":"顶部 Hero 用左右分栏：左为叙事斜线注册视觉，右为 before→after 的 token 计量带；下方每个 section 横向铺开一条带，箭头/括弧符号悬挂在带沿；底部 Ledger 做横向双列（节省数+成本归因）。","mobileTransformation":"390px 下取消斜线分栏，注册视觉旋转为纵向挖斗+回抽竖带；section 由横带改为纵向段落，括弧符号回到行首；计量带改贴左缘的竖向 token 刻度，保留收缩↔回抽的方向感。","mediaStrategy":"无本地合格素材，不加载远程图片；标志视觉全为证据驱动（压缩方向、括弧、检索箭头、token 刻度）的排版几何，不伪造产品截图。","forbiddenPatterns":["三卡指标墙","浏览器/终端/编辑器外壳","装饰性网格与粒子光晕"],"compositionFingerprint":"register-throttle：同调 token 的压缩斜角 + 括弧减法 + 检索回抽箭头，构成横向方空的变量带连续体"},{"id":"dir-b-mnemonic-strata","name":"The Mnemonic Strata · 记忆地层","projectMetaphor":"上下文是被反复压实的记忆地层：SQLite 把每个会话事件像岩层一样写入，压缩截断表层，检索像钻探一样把相关的旧层抽回地表。整页是一张注意力的地质剖面——深层被压低，检索线索如地层间的矿脉回抽。","signatureVisual":"剖面地层图：一列横向切片（每层是一条会话事件带，宽度代表其信息量），下层的带被压缩黑云压薄；一条带刻痕（检索线索，以细线矿脉样式）从最底层蜿蜒回升到顶层，象征 BM25 抽回相关的旧上下文。","heroGeometry":"垂直堆叠的斜切地层：多层半透明岩层自上而下逐层收窄下压，最上层留出天空（当前窗口），最下层最埋没，一条回抽线索贯穿全幅——竖向分层 + 单线攀回，构成越深越薄、一线可回的剖面。","sectionSequence":["The Strata · 会话如何被压实成记忆","The Index · FTS5 为事件编号","The Drill · BM25 把相关回抽","The Keep · 会话连续性 / 持久记忆","The Route · 17 平台统一路由","The Cost · 节省与 FinOps 归因"],"surfaceModel":"地层切片表面：以逐层加深的半透明横带作底，带与带间用不等宽的岩质间歇；切片的透明度、宽度、压薄程度各不相同，体现越深越压缩；检索线单独一层，悬浮在切面之上，不混入切片。","palette":"取岩芯冷调：墨蓝深底（#0c1220），地层用低饱和灰蓝分层（#1d2b42→#16202f），矿脉线用磷光金（#f4c95d）呼应检索的挖掘，压薄黑云用近黑的深靛（#111827）；不使用网格或彩色光晕。","typography":"无衬线标题（Neue / Instrument Serif 风格的微人文）搭配大比例等宽编号（Fira Code）作为层号；用收敛的层级编号（01→06）把每层标记为岩芯刻度；正文系统无衬线，行距松弛以衬层。","desktopComposition":"Hero 为左侧地层剖面（多层斜切+一条攀回金线），右侧承叙事与刻度编号；下方每个 section 以一层岩芯的方式横向展开，检索线从剖面延续下来贯穿各层；Cost 以收束的一层做收尾。","mobileTransformation":"390px 下地层改为单列纵向堆叠，斜切保留但降为扁平横线；检索攀回线改为竖向点线从底部回升贴左缘；穿插的岩层间歇即可换行，编号刻度转贴行首。","mediaStrategy":"无本地素材，不伪造截图；标志视觉为地层剖面 + 单线钻探回抽的纯粹排版几何，切片的宽度/透明度即证据（信息量与压缩度），不用实物图片。","forbiddenPatterns":["三卡指标墙","功能卡片墙","装饰性粒子与弥散光晕"],"compositionFingerprint":"mnemonic-strata：逐层压薄的岩层横带 + 一条贯穿回抽的金线 + 收敛的层号刻度，形成竖向剖面的记忆剖面连续体"},{"id":"dir-c-watershed-fan","name":"The Watershed Fan · 分流闸·流量计","projectMetaphor":"上下文是一股水流：原始输出进入沙箱闸门被节流净化，相关片段存入蓄水库（SQLite），再按需求闸向 17 条平台渠。整页是一个两端不对称的流量分流闸：收窄入，扇出离。","signatureVisual":"分流扇结构：一条粗 token 流（MCP 工具的调用序列）先穿过一个收窄的沙箱闸（变窄的流渠），蓄入一个方形蓄水库，再在右缘扇形展开成 17 条等宽渠线；扇的展开角与渠线数量即路由粒度的证据。","heroGeometry":"非对称扇形构图：左侧一条流入带向右收窄成闸咽喉，咽喉后方扩张为扇形（17 条发散渠线），扇心偏左、扇角大张——收窄→蓄→扇开的单一动量贯穿全幅。","sectionSequence":["The Flow · 原始输出的洪流","The Gate · 沙箱闸节流净化","The Reservoir · SQLite 蓄存相关片段","The Lanes · 17 平台路由分流","The Meter · 节省与 FinOps 计量"],"surfaceModel":"流动渠道表面：以单一流渠为骨架，渠道宽度即数据量，闸口处收窄、扇口处展宽；渠道用实色填充与等宽间隙表达分流，闸与蓄水库用更强的对比色块框定；无卡片、无叠层。","palette":"以流水冷调为主：墨蓝底（#0a0f18），主流道用青绿渐变（#27d0b0→#1f9d8a）表达净水，沙箱闸用高对比的柠檬黄（#f5d547）标定阈值，蓄水库用冷蓝（#5b9dff），fan 渠线用浅冷灰（#8fa7c3）。","typography":"标题用极窄的几何无衬线（如字号 700 的 Space Grotesk）与大留白，把流渠衬得更宽；渠线与平台名用等宽（Fira Mono）作标签；正文系统无衬线，行距中。","desktopComposition":"Hero：左侧叙事 + 右侧完整流入→闸→蓄→扇开分流图；下方每个 section 以渠道段落逐段展开闸/蓄/渠/Meter；阀门与蓄水库保持同一流渠坐标系横向排布。","mobileTransformation":"390px 下扇状改纵向瀑布：流入带自上而下，闸口竖向收窄，蓄水库为横向方块，扇口改为贴底纵向展开的 17 条渠线（竖排）；方向感由右扇旋转为下扇但保留收窄→展开同一动量。","mediaStrategy":"无本地素材，不伪造截图；标志视觉为流渠+闸+蓄水库+扇开渠线的矢量排版几何，渠线数量代表 17 平台路由（证据），闸口表示沙箱净化（证据）。","forbiddenPatterns":["三卡指标墙","终端/编辑器外壳","装饰性径向光晕与网格粒子"],"compositionFingerprint":"watershed-fan：单一流渠的收窄闸口 + 方形蓄水库 + 右缘/底部 17 渠扇开，构成收窄→蓄→扇开的分流闸连续体"}],"selectedDirectionId":"dir-b-mnemonic-strata","selectionRationale":"三个方向都根植于证据（98% 缩减、SQLite/FTS5+BM25 检索、17 平台路由、11 工具、FinOps 归因），但 dir-b 最能表达项目机制本身：上下文优化的本质是把原始数据埋下去（压缩/排除窗）、把相关数据抽回来（检索/持久记忆）。dir-a 偏重单一压缩度量，dir-c 偏重路由分流，而 dir-b 的记忆地层+一线回抽同时覆盖沙箱排出（压薄埋没）、持久记忆（SQLite 岩层）、检索回抽（BM25 金线）与连续性，并天然承接 FinOps 的计量被压缩的部分；其竖向剖面构图在 390px 下可自然重组（地层堆叠+贴左攀回线），桌面/移动概念一致而空间决策不同，且避开了三卡指标墙、卡片墙、终端壳、光晕粒子等禁忌，签名视觉绝不与常见营销 hero 雷同。"},"designBrief":{"schemaVersion":"dodot-design-brief-v1","selectedDirectionId":"dir-b-mnemonic-strata","projectMetaphor":"上下文是被反复压实的记忆地层：SQLite 把每个会话事件像岩层一样写入，压缩截断表层，检索像钻探一样把相关的旧层抽回地表。整页是一张注意力的地质剖面——深层被压低，检索线索如地层间的矿脉回抽。","signatureVisual":"剖面地层图：一列横向切片（每层是一条会话事件带，宽度代表其信息量），下层的带被压缩黑云压薄；一条带刻痕（检索线索，以细线矿脉样式）从最底层蜿蜒回升到顶层，象征 BM25 抽回相关的旧上下文。","compositionFingerprint":"mnemonic-strata：逐层压薄的岩层横带 + 一条贯穿回抽的金线 + 收敛的层号刻度，形成竖向剖面的记忆剖面连续体","heroGeometry":"垂直堆叠的斜切地层：多层半透明岩层自上而下逐层收窄下压，最上层留出天空（当前窗口），最下层最埋没，一条回抽线索贯穿全幅——竖向分层 + 单线攀回，构成越深越薄、一线可回的剖面。","sectionSequence":["The Strata · 会话如何被压实成记忆","The Index · FTS5 为事件编号","The Drill · BM25 把相关回抽","The Keep · 会话连续性 / 持久记忆","The Route · 17 平台统一路由","The Cost · 节省与 FinOps 归因"],"surfaceModel":"地层切片表面：以逐层加深的半透明横带作底，带与带间用不等宽的岩质间歇；切片的透明度、宽度、压薄程度各不相同，体现越深越压缩；检索线单独一层，悬浮在切面之上，不混入切片。","palette":"取岩芯冷调：墨蓝深底（#0c1220），地层用低饱和灰蓝分层（#1d2b42→#16202f），矿脉线用磷光金（#f4c95d）呼应检索的挖掘，压薄黑云用近黑的深靛（#111827）；不使用网格或彩色光晕。","typography":"无衬线标题（Neue / Instrument Serif 风格的微人文）搭配大比例等宽编号（Fira Code）作为层号；用收敛的层级编号（01→06）把每层标记为岩芯刻度；正文系统无衬线，行距松弛以衬层。","desktopComposition":"Hero 为左侧地层剖面（多层斜切+一条攀回金线），右侧承叙事与刻度编号；下方每个 section 以一层岩芯的方式横向展开，检索线从剖面延续下来贯穿各层；Cost 以收束的一层做收尾。","mobileTransformation":"390px 下地层改为单列纵向堆叠，斜切保留但降为扁平横线；检索攀回线改为竖向点线从底部回升贴左缘；穿插的岩层间歇即可换行，编号刻度转贴行首。","mediaStrategy":"无本地素材，不伪造截图；标志视觉为地层剖面 + 单线钻探回抽的纯粹排版几何，切片的宽度/透明度即证据（信息量与压缩度），不用实物图片。","evidencePlan":["以逐层压薄的切片宽度体现 98% 上下文缩减：上层（当前窗口）最宽、最窄层标定压薄，呼应 evidence 中 sandbox 将原始数据排除在窗外。","以贯穿全幅的单条回抽金线 + 层号刻度表达 SQLite/FTS5 索引事件 + BM25 搜索在压缩后回抽，覆盖 claim 中的检索机制。","以岩浆矿脉从底层回升到顶层的走向表达会话连续性 / 持久记忆 / 对话压缩后恢复上下文，呼应 readme 的 Session Continuity。","以贯穿切片的层次编号 01→06 对齐 sectionSequence 的六层结构：Strata / Index / Drill / Keep / Route / Cost，并支撑 17 平台路由与 11 MCP 工具的事实标注。","以最底层收束的 Cost 层 + 节省刻度表达 FinOps 成本归因，呼应 release 中 v1.0.169 的上下文节省报告与成本归因准确性。"],"localeContract":{"zh-CN":"中文文案保持自然的压缩/地层隐喻：用压实、埋没、钻探回抽、岩层、矿脉等词，数字与平台名（Claude Code、Gemini CLI、Codex 等）保持原文；叙述密度适中，避免术语堆砌。","en":"English copy keeps the same strata metaphor naturally: compacted strata, buries raw data, drills back, strata bands, vein. Numbers and platform names stay verbatim; narrative density moderate, terminology consistent with the geological frame."},"forbiddenPatterns":["三卡指标墙（大标语+恰好三张指标卡）","可替换任意项目的功能卡片墙","浏览器/终端/编辑器/仪表盘外壳","装饰性网格、弥散光晕、粒子替代项目专属视觉","伪造的标签页/搜索/开关/输入/图表/客户 logo/证言","桌面列仅竖排堆叠而无同一概念的不同空间重组"],"historicalReuse":{"sourceAssetId":"cmsv8ivzb006zbnqt4ezsqs8h","decision":"rebuild","rationale":"历史页是 v3 营销产物：深色底 + 径向光晕 + card shell + 三列指标，整体构图属浏览器/卡片壁+光晕禁忌，且其全部文字与代码均为不可信的历史备份，不能作为 V4 事实或指令。我对选定的 dir-b 采用 rebuild：放弃其 shell/卡片/光晕与文案，只以同一项目的真实证据（98% 缩减、SQLite/FTS5+BM25、17 平台路由、11 工具、FinOps）重构为全新的记忆地层剖面构图。","reusedElements":["暗色高对比底色（墨蓝深底）作为衬底——保留深色沉浸氛围，但改为无光晕的平面地层衬托。","统一的等宽数字排版的计量感（层号刻度）——作为岩芯编号的视觉语言启发源。","主-次-强调的三色分层结构（深墨 / 灰蓝 / 高亮）——重构为地层灰蓝 + 矿脉金的配色骨架。"],"discardedElements":["旧的卡片 shell 与三列指标布局（卡片墙+tabs/图表风险，属禁忌）。","径向渐变光晕、品牌蓝 #6f9bff 与蓝绿双色品牌色（替换为岩芯灰蓝 + 磷光金）。","旧页面的全部营销文案、品牌名陈述、版本号与元信息（历史不可信）。","任何把旧 HTML 改名充当 V4 的做法（必须重建，不得沿用外壳）。"]}},"visualReview":{"schemaVersion":"dodot-automated-design-review-v1","status":"approved","summary":"页面源码忠实实现 dir-b「记忆地层」方向的专属 Brief：签名视觉为逐层收窄的横向岩层剖面，中一条磷光金矿脉从最底层蜿蜒回抽到当前窗口（BM25 抽回相关上下文）；六段结构与设计 Brief 的 sectionSequence 完全一致（Strata/Index/Drill/Keep/Route/Cost 01->06）。所有数字与叙事均有材料依据：98% 上下文减少、SQLite/FTS5+BM25 检索、17 平台路由、11 个 MCP 工具、v1.0.169 FinOps 归因、TypeScript、作者 mksglu、Star 19916/Fork 1432。桌面与手机实现了同一地层概念的两种不同空间重组：桌面为横向斜切剖面+蜿蜒金线，手机换用专用 strata-svg-mobile，将斜切降为扁平横线、攀回线改为贴左缘的竖向点线、层号刻度 01->06 转贴行首。双语章节一一对等且措辞自然。无浏览器/终端外壳、无装饰性渐变网格/光晕、无泛用三卡墙或卡片墙、无假控件。七项检查全部通过，无 blocking issue，故判定 approved。","checks":{"projectFit":{"passed":true,"evidence":"六段结构与 design-brief.json sectionSequence 完全对齐（The Strata 01 / The Index 02 / The Drill 03 / The Keep 04 / The Route 05 / The Cost 06）。h1「把上下文压实成记忆地层 / Compacting context into memory strata」直接对应方向名「The Mnemonic Strata · 记忆地层」；one-line「为 AI 编码代理优化上下文窗口的 MCP 服务器」与 material.description 的 project projection 一致。正文通篇使用专属地质隐喻（压实/埋没/钻探回抽/岩层/矿脉/岩芯刻度），未回退到通用营销话术。"},"evidenceBound":{"passed":true,"evidence":"关键事实均有材料依据：98% 上下文减少、沙箱把原始数据排除在窗外（material.description）；SQLite+FTS5 索引事件、BM25 搜索压缩后回抽（summary claim / readme）；17 平台路由、11 个 MCP 工具（summary claim）；会话持久记忆（readme Session Continuity）；v1.0.169 改进节省报告与 FinOps 归因（release claim / summary claim）；TypeScript、作者 mksglu、Star 19916、Fork 1432（material.repository）。未发现无依据的虚构指标或捏造数值。"},"signatureVisual":{"passed":true,"evidence":"桌面 hero 的 .strata-svg 真实实现了 brief 定义的「剖面地层图」：6 条横向切片自 01 current window 到 06 FinOps 逐层收窄（x:0->40->80->120->150 边界收缩），中下层用 #111827/#0d131f 压薄；一条磷光金线（stroke #f4c95d, stroke-width 2.5）以贝塞尔曲线从最底层 (185,406) 蜿蜒回升到顶层 (400,120)，两端带圆点，象征 BM25 把相关的旧上下文抽回顶层。配色严格符合 dir-b palette：#0c1220 墨蓝底、#1d2b42->#16202f->#24344e 灰蓝地层、#111827 压薄、#f4c95d 矿脉金；无光晕无装饰网格，是代码中的真实排版几何而非截图。"},"mobileTransformation":{"passed":true,"evidence":"@media (max-width:760px) 未简单堆叠：除 .hero-inner/.two-col 改单列外，专门隐藏桌面 .strata-svg 并启用独立的 .strata-svg-mobile。移动 SVG（viewBox 350x560）对签名视觉做了真正的空间重组：斜切降为扁平横线（横向 rect），攀回线改为贴左缘 x=20 的竖向点线（stroke-dasharray 2 6）自底部 (20,438) 回升到顶部 (20,60)，层号刻度 01->06 转贴行首排于各层左侧，保留收窄->回抽的涌动方向。概念与桌面一致、空间决策不同，消除了 forbiddenPattern 的「桌面列仅竖排堆叠而无同一概念不同空间重组」风险，满足 brief.mobileTransformation 全部要求。"},"localeParity":{"passed":true,"evidence":"zh-CN 与 en 结构一一对应：h1/h2/SVG 文本（01 current window 到 06 FinOps）、hero-badges、two-col 列表、note 注记、cost-row 四格、footer 链接。所有数字（98%/17/11/v1.0.169/19916/1432）与平台名（Claude Code/Gemini CLI/Codex）逐字一致。中文用「压实/岩层/矿脉/钻探回抽/埋没区」等自然隐喻，英文用 strata/vein/buried beds/drill back 同构表达，符合 brief.localeContract，无术语堆砌、无缺译或生硬直译。"},"antiTemplate":{"passed":true,"evidence":"首屏非通用 Hero 标语+三卡：为自定义地层剖面 SVG + 叙事双栏，配真实 CTA 与带数据的 meta。六个 section 均为层带式叙事（sec-no 岩芯刻度 + two-col 列表），而非可替换任意项目的功能卡片墙；Cost 为证据导向的四格 Ledger（98%/17/11/FinOps v1.0.169），对应 evidencePlan 的收尾设计。与三方向中唯一被选定且需 rebuild 的 dir-b 特有概念一致，未落入三卡墙/卡片墙/浏览器终端壳/装饰性渐变网格等禁忌。"},"noFakeUi":{"passed":true,"evidence":"源码仅含真实链接（https://context-mode.com/ 与 https://github.com/mksglu/context-mode），无伪造的标签页/搜索框/开关/输入框/图表/客户 logo/证言。qa-report.json 四个 viewport（zh/en x 桌面/移动）均报 fakeControlCount:0、emptyLinkCount:0、brokenImageCount:0、consoleErrors:[]、networkRequests:[]，且页面未引用任何本地或远程图片素材（mediaStrategy 的纯排版几何已落实）。"}},"issues":[]},"repairEvidence":[{"attempt":1,"before":{"qaPassed":true,"designReviewApproved":false,"qaEvidence":{"zh-CN":[{"viewport":"desktop","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":1716,"cjkCount":1042,"latinLetterCount":294,"scrollWidth":1280,"clientWidth":1280,"scrollHeight":3462,"h1Top":174.921875,"bottomInset":209.640625,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"a76583285884e91f6c88d972574641743f720d5e4300455e41c0ecd5976455eb","middle":"ffbb85da5d6d549db201a6f629596474ec050862ba314bc9529ea55eb0cb2861","bottom":"28bd360329e9a8803d4036826bfed55dac74071e00f3e93e1c606e84e99b0ac2"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":1716,"cjkCount":1042,"latinLetterCount":294,"scrollWidth":390,"clientWidth":390,"scrollHeight":5300,"h1Top":120,"bottomInset":267.078125,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"669c457fab475be2c996a0324ea6293c9ec627de1fd444a402aee905783950ae","middle":"ada6e204ecc9b9d81383366cf6389569e4a03867d9923046344cf7c10b1216ec","bottom":"ead3fd76fe995a5ef9227a70c2b60004de2d5b6466dbb5e801d448f894935753"},"consoleErrors":[],"networkRequests":[]}],"en":[{"viewport":"desktop","passed":true,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":4353,"cjkCount":0,"latinLetterCount":3451,"scrollWidth":1280,"clientWidth":1280,"scrollHeight":3795,"h1Top":136,"bottomInset":210.109375,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"0458aa25ad97a8fce97d51e45a942f9d6babdf9a46df25219d3ee77813c01cbb","middle":"28a5dda929dc3136e19e08280ec1fc9860fa21b1cdbc955cfe464509676d0065","bottom":"d3f54fd1eae5ad80647076dfa37572423040c58ae720307f5257b48e53f14a94"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":4353,"cjkCount":0,"latinLetterCount":3451,"scrollWidth":390,"clientWidth":390,"scrollHeight":6103,"h1Top":120,"bottomInset":267.34375,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"5f34ee55db36fe024541532bdcf06a389e558e26a5d5ab6d7ccb89f479d60895","middle":"46a1ef95a397ebf9c4a21f3cb20cae5e7ccda68f8f8eba7914d97ddda8d10447","bottom":"00128b1b5bcc5982f54b3cb278bcb51e19ca76eef717fa6c5c59e6f7f2703944"},"consoleErrors":[],"networkRequests":[]}]},"visualReview":{"schemaVersion":"dodot-automated-design-review-v1","status":"revise","summary":"页面忠实实现了 dir-b「记忆地层」方向：签名视觉为逐层收窄的横向岩层剖面 + 单条磷光金检索线从底层蜿蜒回抽到顶层，六段结构与 sectionSequence（Strata/Index/Drill/Keep/Route/Cost）一致；98% 削减、SQLite/FTS5+BM25、17 平台、11 工具、v1.0.169、Star/Fork 等数字与叙事均有 material 依据；双语章节结构与措辞自然对等；无假控件、无浏览器外壳、无泛用三卡墙。唯一不通过项为 mobileTransformation：390px 断点仅把网格堆叠为单列并把签名图排到文字下方，未按 brief 要求把剖面重组为贴左攀回的竖向点线、编号刻度贴行首、斜切降为扁平横线——即「桌面列竖排堆叠而无同一概念的不同空间重组」，因此必须 revise。","checks":{"projectFit":{"passed":true,"evidence":"六段结构与 design-brief sectionSequence 完全对齐（Strata/Index/Drill/Keep/Route/Cost，编号 01->06）；h1「把上下文压实成记忆地层 / Compacting context into memory strata」、one-line「为 AI 编码代理优化上下文窗口的 MCP 服务器」与项目定位一致；正文通篇使用压实/埋没/钻探回抽/岩层/矿脉的专属隐喻，未回退到通用营销话术。"},"evidenceBound":{"passed":true,"evidence":"正文与各项数字均有材料依据：98% 上下文减少（material description/summary）、沙箱排除原始数据、SQLite+FTS5 索引、BM25 检索、17 平台路由、11 个 MCP 工具、会话持久记忆、v1.0.169 FinOps 归因、TypeScript、作者 mksglu、Star 19916/Fork 1432、仓库存档页面 two-col & cost 均源自 material.json；未引入无依据的虚构指标。"},"signatureVisual":{"passed":true,"evidence":"hero SVG 实现「剖面地层图」：6 条横向切片自 01 当前窗口到 06 FinOps 逐层收窄（520->40 宽 → 220 宽），下层用 #111827/#0d131f 压薄黑云；单条磷光金线 stroke #f4c95d 以贝塞尔曲线从最底层 (185,406) 蜿蜒回抽到顶层 (400,120)，两端带圆点，对应 BM25 回抽相关旧上下文；配色严格符合 palte（#0c1220 底、#1d2b42->#16202f 灰蓝地层、#111827 压薄、#f4c95d 矿脉），无光晕无网格。"},"mobileTransformation":{"passed":false,"evidence":"唯一 > 760px 断点仅做：.hero-inner 由双栏改 1fr、.strata-wrap 置 order:2、.two-col 改为单列。签名地层 SVG 在移动端原样保留，仅被排到文字之下；brief 明确要求「检索攀回线改为竖向点线从底部回升贴左缘、斜切保留但降为扁平横线、编号刻度转贴行首」，实现中均无相应重组。这属于 brief forbiddenPattern #6「桌面列仅竖排堆叠而无同一概念的不同空间重组」。"},"localeParity":{"passed":true,"evidence":"zh-CN 与 en 结构一一对应：h1/h2/SVG 文本（01 current window->06 FinOps）、hero-badges、two-col 列表、note、cost-row 四格、footer、链接与数字（98%/17/11/v1.0.169/19916/1432）完全一致；英文为自然地质隐喻（strata/vein/buried beds/drills back），中文为「压实/岩层/矿脉/钻探回抽」，无术语堆砌、无缺译。"},"antiTemplate":{"passed":true,"evidence":"非通用 Hero/三卡墙/卡片墙：首屏为自定义地层剖面 SVG + 叙事双栏；六个 section 均为层带式叙事（sec-no 岩芯刻度 + two-col），而非可替换任意项目的功能卡片墙；Cost 为证据导向的四格 Ledger（节省 98%/17 平台/11 工具/FinOps v1.0.169），符合 evidencePlan 的 Cost 收尾设计，非营销卡墙。"},"noFakeUi":{"passed":true,"evidence":"源码仅含真实链接（context-mode.com、github.com/mksglu/context-mode），无伪造标签页/搜索/开关/输入/图表/客户 logo/证言；qa-report 四个 viewport 均 fakeControlCount:0、networkRequests:[]、brokenImageCount:0、consoleErrors:[]，且未引用任何本地/远程图片素材。"}},"issues":[{"code":"mobile-recomposition","severity":"blocking","locale":"all","description":"手机 CSS 仅有单列堆叠，未按 brief mobileTransformation 重组签名视觉。@media (max-width:760px) 只把 .hero-inner/.two-col 改为 1 列、.strata-wrap order:2，地层剖面 SVG 原样保留在文字下方，未形成「同一概念的不同空间重组」；违反 brief 对 dir-b 的移动端要求（攀回线改为贴左缘竖向点线、斜切降为扁平横线、编号刻度转贴行首 01->06）。","repairInstruction":"在窄屏断点内对签名视觉做空间重组而非仅堆叠：可将地层带到竖向叠加并让剖面图改为左侧贴缘的竖向点线攀回（stroke-dasharray 的金线纵排），层号刻度 01->06 随行布局转贴行首，斜切降为扁平横线；保留收窄->回抽的涌动方向，确保 390px 下概念不变而空间决策与桌面不同，从而消除「桌面列仅竖排堆叠」的禁忌。"}]}},"after":{"qaPassed":true,"designReviewApproved":true,"qaEvidence":{"zh-CN":[{"viewport":"desktop","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":1716,"cjkCount":1042,"latinLetterCount":294,"scrollWidth":1280,"clientWidth":1280,"scrollHeight":3462,"h1Top":174.921875,"bottomInset":209.640625,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"a76583285884e91f6c88d972574641743f720d5e4300455e41c0ecd5976455eb","middle":"ffbb85da5d6d549db201a6f629596474ec050862ba314bc9529ea55eb0cb2861","bottom":"28bd360329e9a8803d4036826bfed55dac74071e00f3e93e1c606e84e99b0ac2"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":1716,"cjkCount":1042,"latinLetterCount":294,"scrollWidth":390,"clientWidth":390,"scrollHeight":5483,"h1Top":120,"bottomInset":267,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"23334a55e0ffe150fe896b672b92693867b797132cea0b04783699edc9c4ad81","middle":"033691c38f29e8a5b0d2d64e69bca9d8a9a97e47a449e445e3ca0d12e5694cba","bottom":"842a6602691b7b4999376e225431c224d71a1f611357f0bde411f861eada607c"},"consoleErrors":[],"networkRequests":[]}],"en":[{"viewport":"desktop","passed":true,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":4353,"cjkCount":0,"latinLetterCount":3451,"scrollWidth":1280,"clientWidth":1280,"scrollHeight":3795,"h1Top":136,"bottomInset":210.109375,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"0458aa25ad97a8fce97d51e45a942f9d6babdf9a46df25219d3ee77813c01cbb","middle":"28a5dda929dc3136e19e08280ec1fc9860fa21b1cdbc955cfe464509676d0065","bottom":"d3f54fd1eae5ad80647076dfa37572423040c58ae720307f5257b48e53f14a94"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":4353,"cjkCount":0,"latinLetterCount":3451,"scrollWidth":390,"clientWidth":390,"scrollHeight":6286,"h1Top":120,"bottomInset":267.265625,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"6d3a33087dff950d48a865da09e299bd6829afadf9aadcb2beb295a609f1fb0f","middle":"b315051eaec19b0f685758b6abb503c3a7b31439b31d3b951eb1f25744494288","bottom":"c7753c5e8542e8bffe26cec67a39981d912c90799c32cc89256896b3ffd11941"},"consoleErrors":[],"networkRequests":[]}]},"visualReview":{"schemaVersion":"dodot-automated-design-review-v1","status":"approved","summary":"页面源码忠实实现 dir-b「记忆地层」方向的专属 Brief：签名视觉为逐层收窄的横向岩层剖面，中一条磷光金矿脉从最底层蜿蜒回抽到当前窗口（BM25 抽回相关上下文）；六段结构与设计 Brief 的 sectionSequence 完全一致（Strata/Index/Drill/Keep/Route/Cost 01->06）。所有数字与叙事均有材料依据：98% 上下文减少、SQLite/FTS5+BM25 检索、17 平台路由、11 个 MCP 工具、v1.0.169 FinOps 归因、TypeScript、作者 mksglu、Star 19916/Fork 1432。桌面与手机实现了同一地层概念的两种不同空间重组：桌面为横向斜切剖面+蜿蜒金线，手机换用专用 strata-svg-mobile，将斜切降为扁平横线、攀回线改为贴左缘的竖向点线、层号刻度 01->06 转贴行首。双语章节一一对等且措辞自然。无浏览器/终端外壳、无装饰性渐变网格/光晕、无泛用三卡墙或卡片墙、无假控件。七项检查全部通过，无 blocking issue，故判定 approved。","checks":{"projectFit":{"passed":true,"evidence":"六段结构与 design-brief.json sectionSequence 完全对齐（The Strata 01 / The Index 02 / The Drill 03 / The Keep 04 / The Route 05 / The Cost 06）。h1「把上下文压实成记忆地层 / Compacting context into memory strata」直接对应方向名「The Mnemonic Strata · 记忆地层」；one-line「为 AI 编码代理优化上下文窗口的 MCP 服务器」与 material.description 的 project projection 一致。正文通篇使用专属地质隐喻（压实/埋没/钻探回抽/岩层/矿脉/岩芯刻度），未回退到通用营销话术。"},"evidenceBound":{"passed":true,"evidence":"关键事实均有材料依据：98% 上下文减少、沙箱把原始数据排除在窗外（material.description）；SQLite+FTS5 索引事件、BM25 搜索压缩后回抽（summary claim / readme）；17 平台路由、11 个 MCP 工具（summary claim）；会话持久记忆（readme Session Continuity）；v1.0.169 改进节省报告与 FinOps 归因（release claim / summary claim）；TypeScript、作者 mksglu、Star 19916、Fork 1432（material.repository）。未发现无依据的虚构指标或捏造数值。"},"signatureVisual":{"passed":true,"evidence":"桌面 hero 的 .strata-svg 真实实现了 brief 定义的「剖面地层图」：6 条横向切片自 01 current window 到 06 FinOps 逐层收窄（x:0->40->80->120->150 边界收缩），中下层用 #111827/#0d131f 压薄；一条磷光金线（stroke #f4c95d, stroke-width 2.5）以贝塞尔曲线从最底层 (185,406) 蜿蜒回升到顶层 (400,120)，两端带圆点，象征 BM25 把相关的旧上下文抽回顶层。配色严格符合 dir-b palette：#0c1220 墨蓝底、#1d2b42->#16202f->#24344e 灰蓝地层、#111827 压薄、#f4c95d 矿脉金；无光晕无装饰网格，是代码中的真实排版几何而非截图。"},"mobileTransformation":{"passed":true,"evidence":"@media (max-width:760px) 未简单堆叠：除 .hero-inner/.two-col 改单列外，专门隐藏桌面 .strata-svg 并启用独立的 .strata-svg-mobile。移动 SVG（viewBox 350x560）对签名视觉做了真正的空间重组：斜切降为扁平横线（横向 rect），攀回线改为贴左缘 x=20 的竖向点线（stroke-dasharray 2 6）自底部 (20,438) 回升到顶部 (20,60)，层号刻度 01->06 转贴行首排于各层左侧，保留收窄->回抽的涌动方向。概念与桌面一致、空间决策不同，消除了 forbiddenPattern 的「桌面列仅竖排堆叠而无同一概念不同空间重组」风险，满足 brief.mobileTransformation 全部要求。"},"localeParity":{"passed":true,"evidence":"zh-CN 与 en 结构一一对应：h1/h2/SVG 文本（01 current window 到 06 FinOps）、hero-badges、two-col 列表、note 注记、cost-row 四格、footer 链接。所有数字（98%/17/11/v1.0.169/19916/1432）与平台名（Claude Code/Gemini CLI/Codex）逐字一致。中文用「压实/岩层/矿脉/钻探回抽/埋没区」等自然隐喻，英文用 strata/vein/buried beds/drill back 同构表达，符合 brief.localeContract，无术语堆砌、无缺译或生硬直译。"},"antiTemplate":{"passed":true,"evidence":"首屏非通用 Hero 标语+三卡：为自定义地层剖面 SVG + 叙事双栏，配真实 CTA 与带数据的 meta。六个 section 均为层带式叙事（sec-no 岩芯刻度 + two-col 列表），而非可替换任意项目的功能卡片墙；Cost 为证据导向的四格 Ledger（98%/17/11/FinOps v1.0.169），对应 evidencePlan 的收尾设计。与三方向中唯一被选定且需 rebuild 的 dir-b 特有概念一致，未落入三卡墙/卡片墙/浏览器终端壳/装饰性渐变网格等禁忌。"},"noFakeUi":{"passed":true,"evidence":"源码仅含真实链接（https://context-mode.com/ 与 https://github.com/mksglu/context-mode），无伪造的标签页/搜索框/开关/输入框/图表/客户 logo/证言。qa-report.json 四个 viewport（zh/en x 桌面/移动）均报 fakeControlCount:0、emptyLinkCount:0、brokenImageCount:0、consoleErrors:[]、networkRequests:[]，且页面未引用任何本地或远程图片素材（mediaStrategy 的纯排版几何已落实）。"}},"issues":[]}},"changedFiles":["artifact/zh-CN.html","artifact/en.html"]}],"provenance":{"executor":"opencode","model":"deepseek/deepseek-v4-flash","opencodeVersion":"1.1.58","promptVersion":"dodot-showcase-deepseek-v4","inputHash":"c47a5de8ba48fb894a371da7b0782d5665fed3a2c22a17f3d7be3a67b988f717","artifactHash":"37fb3df5bcf645b79bf4759a0198b5f54a2eba9625452a4fb14ee18f78b8b7ca","designSkill":"dodot-showcase-design","historicalSeed":{"source":"backup_database","sourceCanonicalKey":"github:mksglu/context-mode","assetId":"cmsv8ivzb006zbnqt4ezsqs8h","sourceSchemaVersion":"localized-static-html-v3","sourceGeneratorVersion":"opencode-1.1.58:deepseek/deepseek-v4-flash","sourceArtifactHash":"f64f9fba66ec2cd12b04cada03a95593bf1056d2a3179c2830247a72362e16b5","disposition":"validate_and_upgrade","seedPayloadHash":"168e5c4338a20fa9ae856a42fe21b94960e3009d05005793d1eb61fa24582431"}},"qaEvidence":{"zh-CN":[{"viewport":"desktop","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":1716,"cjkCount":1042,"latinLetterCount":294,"scrollWidth":1280,"clientWidth":1280,"scrollHeight":3462,"h1Top":174.921875,"bottomInset":209.640625,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"a76583285884e91f6c88d972574641743f720d5e4300455e41c0ecd5976455eb","middle":"ffbb85da5d6d549db201a6f629596474ec050862ba314bc9529ea55eb0cb2861","bottom":"28bd360329e9a8803d4036826bfed55dac74071e00f3e93e1c606e84e99b0ac2"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"zh-CN","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":1716,"cjkCount":1042,"latinLetterCount":294,"scrollWidth":390,"clientWidth":390,"scrollHeight":5483,"h1Top":120,"bottomInset":267,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"23334a55e0ffe150fe896b672b92693867b797132cea0b04783699edc9c4ad81","middle":"033691c38f29e8a5b0d2d64e69bca9d8a9a97e47a449e445e3ca0d12e5694cba","bottom":"842a6602691b7b4999376e225431c224d71a1f611357f0bde411f861eada607c"},"consoleErrors":[],"networkRequests":[]}],"en":[{"viewport":"desktop","passed":true,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":4353,"cjkCount":0,"latinLetterCount":3451,"scrollWidth":1280,"clientWidth":1280,"scrollHeight":3795,"h1Top":136,"bottomInset":210.109375,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"0458aa25ad97a8fce97d51e45a942f9d6babdf9a46df25219d3ee77813c01cbb","middle":"28a5dda929dc3136e19e08280ec1fc9860fa21b1cdbc955cfe464509676d0065","bottom":"d3f54fd1eae5ad80647076dfa37572423040c58ae720307f5257b48e53f14a94"},"consoleErrors":[],"networkRequests":[]},{"viewport":"mobile","passed":true,"localePassed":true,"metrics":{"language":"en","h1Count":1,"h2Count":6,"sectionCount":6,"textLength":4353,"cjkCount":0,"latinLetterCount":3451,"scrollWidth":390,"clientWidth":390,"scrollHeight":6286,"h1Top":120,"bottomInset":267.265625,"brokenImageCount":0,"missingAltCount":0,"fakeControlCount":0,"emptyLinkCount":0},"screenshotHashes":{"top":"6d3a33087dff950d48a865da09e299bd6829afadf9aadcb2beb295a609f1fb0f","middle":"b315051eaec19b0f685758b6abb503c3a7b31439b31d3b951eb1f25744494288","bottom":"c7753c5e8542e8bffe26cec67a39981d912c90799c32cc89256896b3ffd11941"},"consoleErrors":[],"networkRequests":[]}]}},"generatorVersion":"opencode-1.1.58:deepseek/deepseek-v4-flash:dodot-showcase-deepseek-v4","publishedAt":"2026-08-20T13:42:51.878Z"},"rawExtraction":{"projectRevisionId":"cmt2zh1hf01356271gzmmymo9","projectVersion":1,"sources":[{"canonicalKey":"github:mksglu/context-mode","role":"primary","revisionVersion":1,"url":"https://github.com/mksglu/context-mode"}],"facts":[{"key":"repository_owner","value":"mksglu","evidence":[{"path":"repository.owner","excerpt":"mksglu"}]},{"key":"repository_name","value":"context-mode","evidence":[{"path":"repository.name","excerpt":"context-mode"}]},{"key":"description","value":"Context window optimization for AI coding agents. Sandboxes tool output (98% reduction), persists session memory, and enforces routing across 17 platforms via MCP + hooks.","evidence":[{"path":"repository.description","excerpt":"Context window optimization for AI coding agents. Sandboxes tool output (98% reduction), persists session memory, and enforces routing across 17 platforms via MCP + hooks."}]},{"key":"primary_language","value":"TypeScript","evidence":[{"path":"repository.language","excerpt":"TypeScript"}]},{"key":"topics","value":"antigravity, claude, claude-code, claude-code-hooks, claude-code-plugins, claude-code-skill, codex, codex-cli, context-mode, copilot, cursor-plugin, kiro, mcp, mcp-server, mcp-tools, openclaw, opencode, pi-agent, skills, zed-extension","evidence":[{"path":"repository.topics","excerpt":"antigravity, claude, claude-code, claude-code-hooks, claude-code-plugins, claude-code-skill, codex, codex-cli, context-mode, copilot, cursor-plugin, kiro, mcp, mcp-server, mcp-tools, openclaw, opencode, pi-agent, skills, zed-extension"}]},{"key":"source_count","value":"1","evidence":[{"path":"projectRevision.sources","excerpt":"github:mksglu/context-mode"}]}],"taxonomy":{"category":"development_tools","subcategory":"data_ai_tools","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Context window optimization for AI coding agents. Sandboxes tool output (98% reduction), persists session memory, and enforces routing across 17 platforms via MCP + hooks."}]},"lifecycle":{"stage":"prototype","confidence":0.65,"evidence":[{"path":"readmeExcerpt","excerpt":"# Context Mode **The other half of the context problem.** [![users](https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fcdn.jsdelivr.net%2Fgh%2Fmksglu%2Fcontext-mode%40main%2Fstats.json&query=%24.message&label=users&color=brightgreen)](https://www.npmjs.com/package/context-mode) [![npm](https://img.shields.io/badge/dynamic/json?url=https%3A%2F%2Fcdn.jsdelivr.net%2Fgh%2Fmksglu%2Fcontext-mode%40main%2Fstats.json&query=%24.npm&label=npm&color=blue)](https://www.npmjs.com/package/context-mod"}]},"userIntents":[{"slug":"build_software","label":"构建或改进软件","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Context window optimization for AI coding agents. Sandboxes tool output (98% reduction), persists session memory, and enforces routing across 17 platforms via MCP + hooks."}]},{"slug":"automate_work","label":"自动化工作流程","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Context window optimization for AI coding agents. Sandboxes tool output (98% reduction), persists session memory, and enforces routing across 17 platforms via MCP + hooks."}]},{"slug":"create_content","label":"创作内容或设计","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Context window optimization for AI coding agents. Sandboxes tool output (98% reduction), persists session memory, and enforces routing across 17 platforms via MCP + hooks."}]},{"slug":"analyze_data","label":"分析与理解数据","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Context window optimization for AI coding agents. Sandboxes tool output (98% reduction), persists session memory, and enforces routing across 17 platforms via MCP + hooks."}]},{"slug":"learn_explore","label":"学习与探索","confidence":0.65,"evidence":[{"path":"repository.description","excerpt":"Context window optimization for AI coding agents. Sandboxes tool output (98% reduction), persists session memory, and enforces routing across 17 platforms via MCP + hooks."}]}],"readme":{"headings":["Context Mode","The Problem","How Context Mode Solves It","Install","Option A — Marketplace plugin (recommended once published)","Option B — Manual install (existing path)","Tools","How the Sandbox Works","How the Knowledge Base Works","Ranking: Reciprocal Rank Fusion","Proximity Reranking","Fuzzy Correction","Smart Snippets","TTL Cache","Progressive Throttling","Session Continuity","Platform Compatibility","Routing Enforcement","Utility Commands","Benchmarks"],"commands":["npm install -g context-mode","npm install -g context-mode","npm install -g context-mode","npm install -g context-mode # the plugin's MCP server runs the global binary","npm install -g context-mode","npm install -g context-mode","npm run install:openclaw","npm run install:openclaw -- /path/to/openclaw-state","npm install -g context-mode","npm install -g context-mode","npm install -g context-mode","npm install -g context-mode","npm install -g context-mode # the plugin's MCP server + hooks run the global binary","npm install -g context-mode","npm install -g context-mode","npm install -g context-mode","bun add context-mode # or: npm install context-mode","npm install -g context-mode","npm install -g context-mode"]},"changeSet":[{"field":"authorName","beforeHash":"74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b","afterHash":"d4eb793a6a322d86cc57f1207d3fc7e4be353048b795f7d3bca826f85b7b78d5"},{"field":"description","beforeHash":"74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b","afterHash":"3b82ee7b40ac785a49f4edc3d98416a8dc38f979c519542aeddc5d9ac4cd4eed"},{"field":"experienceUrl","beforeHash":"74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b","afterHash":"7d54f64bb2aabe5f669b8688747f5cf2e01664c05703220a59f9180e2ad88c93"},{"field":"name","beforeHash":"74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b","afterHash":"f684c7982727e60d964665667585ae0996dc4e67e82a0c6e0056d720714017f0"},{"field":"repository","beforeHash":"74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b","afterHash":"ffc22b09747338bee5b20a2b5dea8838a2d99c01e7d366789bd70bc0a4ba5472"},{"field":"sourceCount","beforeHash":"74234e98afe7498fb5daf1f36ac2d78acc339464f950703b8c019892f982b90b","afterHash":"6b86b273ff34fce19d6b804eff5a3f5747ada4eaa22f1d49c01e52ddb7875b4b"}]},"publishedAt":"2026-08-17T14:05:57.102Z","updatedAt":"2026-08-21T13:26:51.889Z","summary":{"id":"cmt2zh1hi01376271psx2sa09","projectType":"product","oneLine":"为 AI 编码代理优化上下文窗口的 MCP 服务器","overview":"Context Mode 是一个 MCP 服务器，通过沙箱工具将原始数据排除在上下文窗口之外，声称可减少 98% 的上下文占用。它使用 SQLite 和 FTS5 索引会话事件，在对话压缩后通过 BM25 搜索恢复相关上下文，并支持 17 个平台（如 Claude Code、Gemini CLI）通过 MCP 和 hooks 强制执行路由。适合希望减少上下文消耗、提升 AI 编码代理效率的开发者。","highlights":["98% 上下文减少","会话记忆持久化","17 平台路由","11 个 MCP 工具"],"bestFor":["使用 AI 编码代理的开发者","需要长会话连续性的团队","优化上下文成本的工程师"],"caveats":["安装配置较复杂，需按平台操作","部分平台需手动复制路由文件","效果依赖模型和工具使用方式"],"categorySuggestion":"development_tools","subcategorySuggestion":"mcp_integrations","lifecycleStages":["idea_requirements","frontend","backend_data","testing_quality"],"audienceIntents":["build","integrate","operate","evaluate"],"confidence":0.85,"evidence":[{"claim":"Context Mode 是一个 MCP 服务器，通过沙箱工具将原始数据排除在上下文窗口之外，声称可减少 98% 的上下文占用。","sourceType":"description","locator":"https://github.com/mksglu/context-mode"},{"claim":"它使用 SQLite 和 FTS5 索引会话事件，通过 BM25 搜索在对话压缩后恢复相关上下文。","sourceType":"readme","locator":"https://github.com/mksglu/context-mode"},{"claim":"支持 17 个平台，包括 Claude Code、Gemini CLI、Codex 等，通过 MCP 和 hooks 强制执行路由。","sourceType":"description","locator":"https://github.com/mksglu/context-mode"},{"claim":"提供 11 个 MCP 工具，包括六个沙箱工具（如 ctx_execute）和五个元工具（如 ctx_stats）。","sourceType":"readme","locator":"https://github.com/mksglu/context-mode"},{"claim":"最新版本 v1.0.169 改进了上下文节省报告和 FinOps 成本归因的准确性。","sourceType":"release","locator":"https://github.com/mksglu/context-mode/releases/tag/v1.0.169"}],"publishedAt":"2026-08-17T14:05:57.102Z"},"github":{"url":"https://github.com/mksglu/context-mode","ownerAvatarUrl":"https://avatars.githubusercontent.com/u/6067714?v=4","starsCount":19916,"forksCount":1432,"language":"TypeScript","syncedAt":"2026-08-17T14:02:01.379Z","license":null,"defaultBranch":"main","archived":false,"disabled":false,"lastChangedAt":"2026-08-17T06:34:50.000Z","missing":false},"tags":[],"updates":[]},"entryPoint":"discovery","claim":{"state":"unclaimed","verified":false},"initialView":"beautified","community":{"viewer":null,"state":{"likesCount":0,"creativeCount":0,"followersCount":0,"commentsCount":0,"liked":false,"creative":false,"followed":false},"comments":{"comments":[],"nextCursor":null}}}]}],"$L24"]