diff --git a/.agents/notes/implemented/bug-fix/2026-08-05-context-meter-blind-to-compaction.i18n.yaml b/.agents/notes/implemented/bug-fix/2026-08-05-context-meter-blind-to-compaction.i18n.yaml new file mode 100644 index 0000000000..3d9fef5b34 --- /dev/null +++ b/.agents/notes/implemented/bug-fix/2026-08-05-context-meter-blind-to-compaction.i18n.yaml @@ -0,0 +1,6 @@ +# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each +# side as of the last confirmed-consistent state. Both languages carry equal authority; +# after editing either side, bring the other along and re-record with: +# pnpm run verify-translation-pairing --write .agents/notes/implemented/bug-fix/2026-08-05-context-meter-blind-to-compaction.md +2026-08-05-context-meter-blind-to-compaction.md: ab39ae4e109f238960fd60de5e5b61075344f525 +2026-08-05-context-meter-blind-to-compaction.zh.md: c93aa509530e48acf906b18ba85bab4c7d355d69 diff --git a/.agents/notes/implemented/bug-fix/2026-08-05-context-meter-blind-to-compaction.md b/.agents/notes/implemented/bug-fix/2026-08-05-context-meter-blind-to-compaction.md new file mode 100644 index 0000000000..ab39ae4e10 --- /dev/null +++ b/.agents/notes/implemented/bug-fix/2026-08-05-context-meter-blind-to-compaction.md @@ -0,0 +1,46 @@ +# Agent Note: the context meter could not see a compaction + +Status: implemented + +English | [中文](2026-08-05-context-meter-blind-to-compaction.zh.md) + +## Problem + +The composer's [context meter](../feature/2026-08-05-composer-context-meter-breakdown.md) took its ring, percentage, and `~used / capacity` header from `contextPressure.pressureTokens`, the newest provider-reported prompt size. That number moves only when a request reports usage, and compaction reports none: `compact-basic` summarizes through a direct `ctx.llm.stream()` call and appends `compact/start`, `compact/summary`, the replacement `user/message`, and `compact/end` — no `assistant/message`, no usage chunk. + +So the meter was frozen across the one action taken to change it. Driving a real `compactNow` through the agent loop: + +``` +BEFORE compact: ring=4% header=~4227/100000 rows=[system 18, tools 0, messages 4365] +AFTER compact: ring=4% header=~4227/100000 rows=[system 18, tools 0, messages 286] +``` + +The composition rows, which fold the surface, dropped by 93%. The ring — the primary affordance, and the reason a user opens the panel right after compacting — did not move at all, and would not until an entire further turn completed. The panel then showed a header and rows disagreeing by more than an order of magnitude, at exactly the moment a reader was most likely to add the rows up. + +## Decision + +`contextPressure` publishes a second numerator, `projectedTokens`: the provider sample plus the heuristic repricing of everything the surface gained or lost since that sample was taken, clamped at zero. The fold carries the priced surface through the shared `surface-fold.ts` and stamps `sampledSurfaceTokens` when a usage sample lands — **before** the same event joins the surface, so an `assistant/message` anchors against the surface its own request actually carried. `stateVersion` moves to 3. + +Only the delta is estimated. The anchor stays provider-exact, which keeps the estimator's systematic CJK and JSON-schema underpricing out of the occupancy figure while still letting the number react the moment content lands or a span is shadowed. `contextOccupancy` reads `projectedTokens` and falls back to the bare sample, so a projection restored from a pre-field checkpoint degrades to the old behavior instead of vanishing. + +This reverses the "the ring, header, and bar length stay provider-exact" half of the [context meter decision](../feature/2026-08-05-composer-context-meter-breakdown.md). What that decision was protecting — not fabricating precision by scaling heuristic rows to a provider total — is preserved: the rows are still unscaled, and the header still does not equal their sum. What changed is the recognition that "provider-exact but describing a request two compactions ago" is not the more truthful figure. + +## Alternatives considered + +**Project `measure().totalTokens` instead.** The measurement service already composes exactly this (`baseline` anchor plus signed `surfaceDeltaTokens`), and it reacts correctly — measured at 4383 → 304 across the same compaction. But it is a service over private replay state, not a pure fold, and a projection cannot call it. Reproducing its anchor inside a `ProjectionDefinition` needs `_estimateProviderAssistant`'s random access to the chunk provenance (`session.events[seq]`), which `apply(state, event)` does not have. Anchoring on the sampled surface total is the same idea reachable from a pure per-event fold. + +**Emit a synthetic usage record at the end of compaction.** Would move `pressureTokens` itself, but the only usage compaction holds is the summarization request's own — a different prompt entirely. Recording it as the conversation's prompt size would be a lie in the durable log rather than in one display. + +**Let the UI subtract, exposing `sampledSurfaceTokens` and reading `contextBreakdown.messageTokens`.** Splits one figure's arithmetic across two projections and the client. The host owns the vocabulary; it should publish the whole value. + +## Consequences + +Occupancy now advances with every surface event rather than once per turn, so the ring creeps up as a turn produces tool results instead of jumping at its end — and drops the instant a compaction lands. That is more projection frames on the wire: one per surface event for `contextPressure`, the rate `contextBreakdown` already ran at. + +The panel's composition rows still do not sum to the header, and now for one clearly-stated reason instead of two: the rows carry the estimator's error, the header's anchor does not. The remaining lever is estimator accuracy (CJK-aware weighting in `estimate.ts`), which changes no seam. + +`sampledSurfaceTokens` assumes nothing joins the surface between a step's request and its usage report. The loop admits steering and context before `buildRequest` and drains tool results after `assistant/message`, so that holds; if it ever stops holding, the error is bounded by one message and self-corrects at the next sample. + +## Testing + +`packages/llm/token-meter/tests/token-usage-projection.spec.ts` covers the carry-forward across surface growth and a compaction (the sample holding still while the projection shrinks) and the zero clamp when heuristic error would drive the figure negative. `packages/client/ui-conversation/tests/context-meter.spec.tsx` pins the ring reading the projected figure, and `chat-stats-bash-sample.spec.tsx` pins `contextOccupancy`'s preference and its fallback. The end-to-end numbers above came from driving `BasicCompactService.compactNow` through a real `AgentLoop` with the projection registry mounted. diff --git a/.agents/notes/implemented/bug-fix/2026-08-05-context-meter-blind-to-compaction.zh.md b/.agents/notes/implemented/bug-fix/2026-08-05-context-meter-blind-to-compaction.zh.md new file mode 100644 index 0000000000..c93aa50953 --- /dev/null +++ b/.agents/notes/implemented/bug-fix/2026-08-05-context-meter-blind-to-compaction.zh.md @@ -0,0 +1,46 @@ +# Agent Note：上下文仪表看不见压缩 + +Status: implemented + +[English](2026-08-05-context-meter-blind-to-compaction.md) | 中文 + +## 问题 + +composer 的[上下文仪表](../feature/2026-08-05-composer-context-meter-breakdown.md)的圆环、百分比与 `~已用 / 容量` 标题都取自 `contextPressure.pressureTokens`，即提供方报告的最新提示词规模。这个数字只在某个请求报告用量时才会移动，而压缩不报告用量：`compact-basic` 通过直连的 `ctx.llm.stream()` 调用生成摘要，只追加 `compact/start`、`compact/summary`、用作替换的 `user/message` 和 `compact/end`——没有 `assistant/message`，也没有用量分片。 + +于是在唯一一个专门用来改变它的操作面前，这块仪表纹丝不动。通过真实 agent loop 驱动一次 `compactNow`： + +``` +BEFORE compact: ring=4% header=~4227/100000 rows=[system 18, tools 0, messages 4365] +AFTER compact: ring=4% header=~4227/100000 rows=[system 18, tools 0, messages 286] +``` + +折叠表层得出的组成明细行下降了 93%。而圆环——那个主要的可操作元素，也正是用户压缩完立刻会去点开面板的理由——完全没动，而且要等到又跑完一整轮才会动。此时面板上的标题与明细行相差一个数量级以上，恰恰发生在读者最可能去把明细行加总的时刻。 + +## 决策 + +`contextPressure` 发布第二个分子 `projectedTokens`：在提供方样本之上，加上自取样以来表层增减部分的启发式重新计价，下界钳制为零。该折叠通过共享的 `surface-fold.ts` 携带已计价的表层，并在用量样本落地时记下 `sampledSurfaceTokens`——记录时机在同一条事件加入表层**之前**，因此 `assistant/message` 锚定的正是它自己那次请求实际携带的表层。`stateVersion` 提升到 3。 + +只有增量部分是估算的。锚点保持提供方精确值，从而把估算器对 CJK 文本与 JSON schema 的系统性低估挡在占用率数字之外，同时又让这个数字能在内容落地或某段区间被遮蔽的瞬间做出反应。`contextOccupancy` 读取 `projectedTokens`，并回退到裸样本，因此从不含该字段的检查点恢复出来的投影会退化为旧行为，而不是直接消失。 + +这推翻了[上下文仪表决策](../feature/2026-08-05-composer-context-meter-breakdown.md)中「圆环、标题与进度条总长保持提供方精确值」的那一半。那条决策真正想守住的东西——不要把启发式明细行按比例缩放到提供方总量、从而伪造精度——依然守住了：明细行仍未被缩放，标题仍不等于它们之和。改变的是这样一个认识：「提供方精确、但描述的是两次压缩之前那个请求」并不是更真实的数字。 + +## 备选方案 + +**改为投影 `measure().totalTokens`。** 测量服务本来就合成了正是这个量（`baseline` 锚点加有符号的 `surfaceDeltaTokens`），而且反应正确——同一次压缩前后实测为 4383 → 304。但它是一个建立在私有重放状态上的服务，不是纯折叠，投影无法调用它。要在 `ProjectionDefinition` 内部复现它的锚点，需要 `_estimateProviderAssistant` 对分片来源的随机访问（`session.events[seq]`），而 `apply(state, event)` 拿不到。以取样时的表层总量作为锚点，是同一个思路在纯逐事件折叠中可达的版本。 + +**在压缩结束时补写一条合成的用量记录。** 这确实能推动 `pressureTokens` 本身，但压缩手上唯一的用量是摘要请求自己的用量——那是完全另一个提示词。把它记成本对话的提示词规模，等于把谎言写进持久日志，而不只是写进某一处展示。 + +**让 UI 自己做减法：暴露 `sampledSurfaceTokens`，再读 `contextBreakdown.messageTokens`。** 这会把一个数字的算术拆散到两个投影和客户端三处。词汇的所有者是宿主，就应当由它发布完整值。 + +## 影响 + +占用率现在随每个表层事件推进，而不再是每轮跳一次，因此一轮中产生工具结果时圆环会持续爬升，而不是等到轮次结束才跳变——压缩落地的瞬间它也会掉下来。代价是线路上多了投影帧：`contextPressure` 每个表层事件推一帧，也就是 `contextBreakdown` 本来就在跑的频率。 + +面板的组成明细行仍然加不出标题数字，但现在只剩一个能讲清楚的原因，而不是两个：明细行带着估算器的误差，标题的锚点不带。剩下的抓手是估算精度（在 `estimate.ts` 里做 CJK 感知加权），它不改动任何 seam。 + +`sampledSurfaceTokens` 依赖一个前提：在某个步骤的请求与它的用量报告之间，不会有新内容加入表层。循环在 `buildRequest` 之前接纳 steering 与 context，在 `assistant/message` 之后才排空工具结果，因此该前提成立；即便将来不再成立，误差也被限制在一条消息以内，并在下一个样本处自行纠正。 + +## 测试 + +`packages/llm/token-meter/tests/token-usage-projection.spec.ts` 覆盖了样本在表层增长与一次压缩上的推进（样本保持不动而投影值缩小），以及启发式误差会把数字压到负数时的零钳制。`packages/client/ui-conversation/tests/context-meter.spec.tsx` 钉住圆环读取投影值这一点，`chat-stats-bash-sample.spec.tsx` 钉住 `contextOccupancy` 的优先级与回退。上面那组端到端数字来自在挂载了投影注册表的真实 `AgentLoop` 上驱动 `BasicCompactService.compactNow`。 diff --git a/.agents/notes/implemented/feature/2026-08-04-web-latency-throughput-metrics.i18n.yaml b/.agents/notes/implemented/feature/2026-08-04-web-latency-throughput-metrics.i18n.yaml new file mode 100644 index 0000000000..16ccf1cff3 --- /dev/null +++ b/.agents/notes/implemented/feature/2026-08-04-web-latency-throughput-metrics.i18n.yaml @@ -0,0 +1,6 @@ +# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each +# side as of the last confirmed-consistent state. Both languages carry equal authority; +# after editing either side, bring the other along and re-record with: +# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-04-web-latency-throughput-metrics.md +2026-08-04-web-latency-throughput-metrics.md: 4d7627a9a38127259f1ba07121cea7282f794f2e +2026-08-04-web-latency-throughput-metrics.zh.md: 09e59242b799a3f4d03b9e0259f1a5a0d85f6a53 diff --git a/.agents/notes/implemented/feature/2026-08-04-web-latency-throughput-metrics.md b/.agents/notes/implemented/feature/2026-08-04-web-latency-throughput-metrics.md new file mode 100644 index 0000000000..4d7627a9a3 --- /dev/null +++ b/.agents/notes/implemented/feature/2026-08-04-web-latency-throughput-metrics.md @@ -0,0 +1,31 @@ +# Agent Note: Web turn and window latency/throughput metrics + +Status: implemented + +English | [中文](2026-08-04-web-latency-throughput-metrics.zh.md) + +## Problem + +The Web chat records per-step LLM timing (`stepStartTime` / `firstTokenTime` / `completedTime`) and per-step usage, and the trajectory view exposes them per step, but the chat surface answers neither "how responsive was this turn" nor "how fast is this session going": the assistant footer shows only the turn wall time, and the stats line folds only wall-time totals. + +## Decision + +A package-local fold, `ui-conversation`'s `chat/turn-metrics.ts`, is the single derivation from assistant nodes to latency/throughput readings. `assistantStepReading` turns one node into a step reading: TTFT needs both `stepStartTime` and `firstTokenTime`, decode span needs `firstTokenTime`, negative spans clamp to zero, and output tokens come from the untrusted `usage` value only when they are finite and non-negative. `deriveTurnMetrics` folds readings per turn: the lowest-numbered step owns the turn's TTFT slot, and throughput divides the summed output tokens by the summed decode spans over exactly the steps carrying both, so an unsampled step drops out instead of skewing the ratio; a turn with neither figure emits no entry. + +The assistant footer appends the readings to the existing hover-revealed time chrome after `Ran for`, as `TTFT {s}s · {tps} tok/s`, each omitted independently when unrecorded. ChatView shows a turn's readings only when that turn's `turnTimings` entry has an `endTime`: the loaded window is a contiguous log suffix, so an in-window settled turn carries every one of its steps and the first-step TTFT is genuine rather than a window artifact. `formatLatencySeconds` is unit-less so each locale template owns its second suffix (`TTFT {seconds}s` / `首 token {seconds}秒`). + +The stats line reuses the same step reading in its window fold: `deriveStats` accumulates TTFT sum/count and decode span/tokens, rendering a latency/throughput group localized through the `conversation` locale namespace (`TTFT avg … · … tok/s` in English) beside the LLM/tool wall times. The turn-count, step-count, duration, cache, and token labels use the same namespace. Like those wall times the group is window-scoped and folds no billing; token accounting stays on the token-meter projections. + +## Alternatives considered + +**A durable session projection (token-meter shape).** A `ProjectionDefinition` folding step timings host-side would survive compaction and window paging and cover the whole log. Deferred, not rejected: projection state must stay O(1) (averages, not percentiles), it needs a host change plus a schema, and the chat stats line is already documented as window-scoped for its duration facts — the new group joins that scope. A later PR can add the durable projection without moving these readings. + +**Per-step footer chrome.** Showing each assistant message its own TTFT would attach chrome to mid-turn narration nodes, which the footer design deliberately keeps chrome-free; the trajectory view already exposes per-step timing detail. + +**Gating footer metrics on node presence instead of `turn/end` timing.** Rendering whatever steps happen to be loaded would show a plausible-looking TTFT that is actually the first *loaded* step after paging. The `endTime` gate plus the suffix-window invariant makes the displayed figure the turn's true first-step latency or nothing. + +## Consequences + +A settled in-window turn's footer reveals `TTFT`/`tok/s` on hover after the wall time, and the stats line shows window-average latency and throughput with localized labels beside its wall times, all without new session events or host changes. Metrics degrade by omission: providers or steps without timing or usage samples drop individual figures rather than rendering zeros. Older history outside the loaded window stays uncounted, recorded in the package README's stats-line limitation. + +Both readings divide by measured wall time, so neither is reproducible: the same replayed scenario yielded 69 and 70 tok/s on consecutive local runs, and a 3 ms replayed stream reads 26333 tok/s. The Web aria goldens therefore normalize throughput to `{{throughput}}` beside the existing `{{duration}}`, and the footer's decorative separators gained flanking spaces — without them the readings concatenate into one accessible string (`Ran for 13sTTFT 0.2s12 tok/s`), which both loses the reading boundaries a screen reader needs and denies `{{duration}}` the word boundary it matches on. diff --git a/.agents/notes/implemented/feature/2026-08-04-web-latency-throughput-metrics.zh.md b/.agents/notes/implemented/feature/2026-08-04-web-latency-throughput-metrics.zh.md new file mode 100644 index 0000000000..09e59242b7 --- /dev/null +++ b/.agents/notes/implemented/feature/2026-08-04-web-latency-throughput-metrics.zh.md @@ -0,0 +1,31 @@ +# Agent Note: Web 轮次与窗口级延迟/吞吐指标 + +Status: implemented + +[English](2026-08-04-web-latency-throughput-metrics.md) | 中文 + +## 问题 + +Web 聊天已经记录了逐步骤的 LLM 计时（`stepStartTime`／`firstTokenTime`／`completedTime`）和逐步骤 usage，trajectory 视图也按步骤展示它们，但聊天界面既回答不了「这一轮响应有多快」，也回答不了「这个会话跑得有多快」：assistant 页脚只显示轮次实际耗时，统计行也只折算墙钟时间总量。 + +## 决策 + +包内折算 `ui-conversation` 的 `chat/turn-metrics.ts` 是从 assistant 节点推导延迟/吞吐读数的唯一位置。`assistantStepReading` 把一个节点转成一次步骤读数：TTFT（首 token 延迟）需要 `stepStartTime` 与 `firstTokenTime` 同时存在，解码时长需要 `firstTokenTime`，负时长收敛为零，输出 token 数只在不可信的 `usage` 值有限且非负时才采纳。`deriveTurnMetrics` 按轮次折算读数：编号最小的步骤拥有该轮次的 TTFT 槽位，吞吐用「同时携带两者的那些步骤」的输出 token 总和除以解码时长总和，因此缺采样的步骤直接退出而不是让比值失真；两个数字都没有的轮次不产生条目。 + +assistant 页脚把读数追加到既有 hover 显示的时间附属元素中、`用时` 之后，形如 `首 token {s}秒 · {tps} tok/s`，未记录的数字各自省略。ChatView 仅在该轮次的 `turnTimings` 条目带有 `endTime` 时才显示读数：已加载窗口是日志的连续后缀，因此窗口内已结算的轮次必然带着它的全部步骤，首步 TTFT 是真实值而非窗口截断的产物。`formatLatencySeconds` 不带单位，各语言模板各自拥有秒后缀（`TTFT {seconds}s`／`首 token {seconds}秒`）。 + +统计行在其窗口折算中复用同一份步骤读数：`deriveStats` 累计 TTFT 总和／计数与解码时长／token 数，在 LLM／工具墙钟时间旁渲染经 `conversation` locale 命名空间本地化的延迟／吞吐分组（中文为 `首 token 平均 … · … tok/s`）。轮次计数、步骤计数、耗时、缓存与 token 各项的标签也使用同一命名空间。与那些墙钟时间一样，该分组是窗口作用域的，不折算任何计费；token 账目仍归 token-meter 投影。 + +## 考虑过的替代方案 + +**持久的会话投影（token-meter 形态）。** 在 host 侧用 `ProjectionDefinition` 折算步骤计时可以跨越压缩与窗口分页、覆盖整个日志。是暂缓而非否决：投影状态必须保持 O(1)（只能均值，不能分位数），它需要 host 改动加 schema，而聊天统计行的耗时事实本就被记录为窗口作用域——新分组沿用该作用域。后续 PR 可以在不挪动这些读数的情况下补上持久投影。 + +**逐步骤页脚附属元素。** 让每条 assistant 消息显示自己的 TTFT，会给轮次中段的叙述节点挂上附属元素，而页脚设计刻意让它们保持无 chrome；trajectory 视图已经暴露逐步骤计时细节。 + +**用节点是否在场而非 `turn/end` 计时做页脚门控。** 直接渲染碰巧加载到的步骤，会展示一个貌似合理、实为分页后「首个已加载步骤」的 TTFT。`endTime` 门控加上后缀窗口不变量，使显示的数字要么是该轮次真实的首步延迟，要么什么都不显示。 + +## 后果 + +窗口内已结算轮次的页脚在 hover 时于实际耗时之后显示 `首 token`／`tok/s`，统计行在墙钟时间旁以本地化标签显示窗口平均延迟与吞吐，全程不新增会话事件、不改 host。指标以省略的方式退化：没有计时或 usage 采样的提供方或步骤只是丢掉对应数字，而不会渲染成零。已加载窗口之外的更早历史仍不计入，已记录在包 README 的统计行限制中。 + +两个读数都以实测墙钟时间作分母，因此都不可复现：同一个回放场景在本机连续两次跑出 69 与 70 tok/s，而一段 3 毫秒的回放流会读成 26333 tok/s。因此 Web aria golden 在既有的 `{{duration}}` 之外，把吞吐归一化为 `{{throughput}}`；页脚的装饰性分隔符也补上了两侧空格——没有它们，这些读数会连成一整串无障碍文本（`Ran for 13sTTFT 0.2s12 tok/s`），既让屏幕阅读器失去读数之间的边界，也让 `{{duration}}` 失去它赖以匹配的词边界。 diff --git a/.agents/notes/implemented/feature/2026-08-05-composer-context-meter-breakdown.i18n.yaml b/.agents/notes/implemented/feature/2026-08-05-composer-context-meter-breakdown.i18n.yaml new file mode 100644 index 0000000000..059f6d2a0e --- /dev/null +++ b/.agents/notes/implemented/feature/2026-08-05-composer-context-meter-breakdown.i18n.yaml @@ -0,0 +1,6 @@ +# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each +# side as of the last confirmed-consistent state. Both languages carry equal authority; +# after editing either side, bring the other along and re-record with: +# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-05-composer-context-meter-breakdown.md +2026-08-05-composer-context-meter-breakdown.md: a757bcbc8bc57f4c3a16f663a9c922155b8bb575 +2026-08-05-composer-context-meter-breakdown.zh.md: 441fd3013f2db721527838955b5ce227865615f1 diff --git a/.agents/notes/implemented/feature/2026-08-05-composer-context-meter-breakdown.md b/.agents/notes/implemented/feature/2026-08-05-composer-context-meter-breakdown.md new file mode 100644 index 0000000000..a757bcbc8b --- /dev/null +++ b/.agents/notes/implemented/feature/2026-08-05-composer-context-meter-breakdown.md @@ -0,0 +1,31 @@ +# Agent Note: Composer context meter with heuristic composition breakdown + +Status: implemented + +English | [中文](2026-08-05-composer-context-meter-breakdown.zh.md) + +## Problem + +The Web chat's stats line showed context occupancy as one inline figure (`Context N% of X`) among its billing groups. That answers "how full" but not "what fills it": nothing showed how the window divides between the system prompt, tool schemas, and conversation, and the one-line row has no room for that detail. The available numbers also live in two vocabularies — the provider-exact billed prompt size from `contextPressure` versus the token-meter's fixed character heuristic — and no existing surface could present composition without conflating them. + +## Decision + +Three cooperating pieces, one per package boundary: + +`dsh-session` exports the pure `deriveEventMessage(event)` (previously reachable only as a `Session` method, which now delegates to it) so a host-side fold can price surface nodes without a `Session` instance. + +`dsh-token-meter` extracts its pricing heuristic into `src/estimate.ts` and its positional surface fold into `src/surface-fold.ts` — both shared verbatim with the measurement service — and registers a third session projection, `contextBreakdown`, carrying `systemTokens` / `toolsTokens` / `messageTokens`. Envelope figures reprice last-wins on each `request/header` through `canonicalHeader`; the message figure replays `foldSurfaceTokens` over a per-node `{seq, tokens}` list, so it equals `measure().surfaceTokens` at every event boundary by construction and compaction shrinks it the way it shrinks the next request. The shared fold is total and allocation-fresh — it returns the next surface rather than mutating one — which keeps the service's validate-before-commit replay transaction intact: a throw leaves the replay cursor unmoved and the same malformed event fails identically on retry. A replace range absent from the folded surface throws: committed logs are surface-validated at append time, so an unresolvable range is log corruption, not a skippable event. + +`ui-conversation` moves context occupancy off the stats line (one home per fact) onto a composer-trailing `ContextMeter`: a 14px occupancy ring after the model seat fed by `contextPressure`, click-opening a panel that pairs the provider-exact percent and `~used / capacity` header with a 4px color-segmented bar and `~`-prefixed composition rows. The two vocabularies deliberately never reconcile — the heuristic shares only proportion the bar's colored segments and rows, each marked `~` because the fixed 4-chars-per-token heuristic systematically underprices CJK text and code. (The ring, header, and bar length were provider-exact as shipped here; they now read the provider-anchored `projectedTokens` instead, because the bare sample could not see a compaction — see [the meter's compaction blindness](../bug-fix/2026-08-05-context-meter-blind-to-compaction.md).) The header is one localized sentence (`context.aria`, shared with the ring's accessible name) split around its `{percent}` slot, so each locale owns the reading's position — English leads with it, Chinese trails it — while the reading keeps its own tone; a bar part whose width computes to zero is dropped rather than rendered, because `.segment`'s min-width would otherwise paint a filled sliver at 0% occupancy. + +## Alternatives considered + +**Deriving composition client-side from the loaded window.** The window is a contiguous log suffix: the `request/header` events carrying the system prompt and tool schemas may sit outside it, and paging would silently change the figures. Only a durable host-side projection survives paging and compaction, which is why the data crosses the wire as a third projection rather than a chat-window fold. + +**Scaling the heuristic rows to sum to `pressureTokens`.** Forced reconciliation fabricates precision: pressure lags one request, includes provider envelope overhead the estimator never models, and would make the rows move when nothing in the composition changed. Showing the estimator's real vocabulary with an explicit `~` was chosen instead. + +**Finer categories (rules, skills, MCP tools) as in Claude Code's `/context`.** Not separable here: the harness folds those contributions into the system text and the tools list before the request header exists, so three categories are the honest resolution. + +## Consequences + +Token-meter now registers three projection keys; unloading removes all three, and `contextBreakdown` restores from JSON checkpoints (`stateVersion` 1). The stats line dropped its Context group and the ring is the sole context UI. The panel's heuristic rows visibly disagree with the provider-exact header — accepted and signposted by the `~` prefix; improving estimate accuracy (for example CJK-aware weighting) is localized to `estimate.ts` and changes no seam. The legend's purple segment tint is a literal color because the design platform ships no purple static token. diff --git a/.agents/notes/implemented/feature/2026-08-05-composer-context-meter-breakdown.zh.md b/.agents/notes/implemented/feature/2026-08-05-composer-context-meter-breakdown.zh.md new file mode 100644 index 0000000000..441fd3013f --- /dev/null +++ b/.agents/notes/implemented/feature/2026-08-05-composer-context-meter-breakdown.zh.md @@ -0,0 +1,31 @@ +# Agent Note: composer 上下文占用圆环与启发式组成明细 + +Status: implemented + +[English](2026-08-05-composer-context-meter-breakdown.md) | 中文 + +## 问题 + +Web 聊天的统计行把上下文占用率作为一个行内数字（`Context N% of X`）挤在计费分组之间。它回答了「有多满」，却回答不了「被什么占满」：没有任何地方展示窗口在系统提示词、工具 schema 与对话之间如何分配，而单行统计行也容纳不下这种明细。可用的数字还分属两套口径——来自 `contextPressure` 的提供方精确计费 prompt 规模，与 token-meter 的固定字符启发式——没有任何既有界面能在不混淆两者的前提下展示组成。 + +## 决定 + +三个协作部分，每个包边界一个： + +`dsh-session` 导出纯函数 `deriveEventMessage(event)`（此前只能通过 `Session` 方法访问，该方法现在委托给它），使 host 侧 fold 无需 `Session` 实例即可为表层节点计价。 + +`dsh-token-meter` 把计价启发式抽取到 `src/estimate.ts`、把位置表层折叠抽取到 `src/surface-fold.ts`（两者都与测量服务逐字共享），并注册第三个会话投影 `contextBreakdown`，携带 `systemTokens` / `toolsTokens` / `messageTokens`。envelope 数字在每条 `request/header` 上经 `canonicalHeader` 按后者胜重新计价；消息数字在逐节点 `{seq, tokens}` 列表上重放 `foldSurfaceTokens`，因此它在每个事件边界上按构造等于 `measure().surfaceTokens`，压缩会像缩小下一个请求那样缩小它。这份共享折叠是全函数且总是新建数组——返回下一个表层而不是原地改写——从而保留了服务侧「先校验再提交」的重放事务：抛出时重放游标不前进，同一条畸形事件在重试时报同样的错。折叠表层中不存在的替换范围会直接抛出：已提交日志在追加时就经过表层校验，无法解析的范围是日志损坏，而不是可跳过的事件。 + +`ui-conversation` 把上下文占用率从统计行移走（一个事实一个家），放到 composer 尾部的 `ContextMeter`：模型座位之后的一枚 14px 占用圆环，由 `contextPressure` 供数，点击弹出的面板把提供方精确的百分比与 `~已用 / 容量` 标题，与 4px 分色分段进度条及带 `~` 前缀的组成明细行并列。两套口径刻意永不对账——启发式占比只用于切分进度条的彩色分段与明细行，且每个启发式数字都标 `~`，因为固定的「4 字符≈1 token」启发式会系统性低估 CJK 文本与代码。（本记录落地时，圆环、标题与进度条总长取的是提供方精确值；它们现在改读锚定在提供方读数上的 `projectedTokens`，因为裸样本看不见压缩——见[仪表对压缩的失明](../bug-fix/2026-08-05-context-meter-blind-to-compaction.md)。）标题是一整句本地化文案（`context.aria`，与圆环的无障碍名共用），在 `{percent}` 槽位处切开渲染，于是读数的位置由各语言自己决定——英文在前、中文在后——同时读数保留自己的字重；宽度算出为零的分段直接不渲染，否则 `.segment` 的 min-width 会在 0% 占用时画出一段填充色。 + +## 备选方案 + +**在客户端从已加载窗口推导组成。** 窗口是日志的连续后缀：携带系统提示词与工具 schema 的 `request/header` 事件可能在窗口之外，翻页还会让数字悄悄变化。只有持久的 host 侧投影能在翻页与压缩后幸存，这正是数据以第三个投影而非聊天窗口 fold 的形式过线的原因。 + +**把启发式明细行按比例缩放到与 `pressureTokens` 相加一致。** 强行对账是在捏造精度：压力滞后一个请求，还包含估算器从不建模的提供方封装开销，会让明细行在组成毫无变化时也跟着变动。最终选择以显式 `~` 展示估算器的真实口径。 + +**更细的类别（rules、skills、MCP 工具，如 Claude Code 的 `/context`）。** 在这里不可分：harness 在请求标头存在之前就把这些贡献折入系统文本与工具列表，因此三个类别是诚实的分辨率。 + +## 后果 + +token-meter 现在注册三个投影键；卸载会移除全部三个，`contextBreakdown` 可从 JSON 检查点恢复（`stateVersion` 为 1）。统计行删除了 Context 分组，圆环成为唯一的上下文 UI。面板的启发式明细行与提供方精确的标题数字肉眼可见地不一致——已接受并以 `~` 前缀标示；提升估算精度（例如按 CJK 加权）只需改动 `estimate.ts`，不涉及任何 seam。图例的紫色分段色值是字面量，因为设计平台没有紫色静态 token。 diff --git a/apps/web/tests/built-boot.snapshot.ts b/apps/web/tests/built-boot.snapshot.ts index 12487d0c43..b102dff8b8 100644 --- a/apps/web/tests/built-boot.snapshot.ts +++ b/apps/web/tests/built-boot.snapshot.ts @@ -121,6 +121,21 @@ it('boots the built plugin graph and renders a fixture session end to end', asyn expect(document.querySelector('[data-sample="bash"]')).not.toBeNull() }, { timeout: 10_000 }) + // Resolve the resident approval so the ordinary composer bar (which owns + // ContextMeter) resumes without replacing the session shell. This minimal + // boot graph intentionally does not mount the separate question UI plugin. + fireEvent.click(await screen.findByRole('button', { name: 'Allow once' })) + + // The fixture mirrors all three token-meter projections, so the assembled + // ContextMeter reaches its composition panel instead of only the occupancy + // fallback path. + const contextTrigger = await screen.findByRole('button', { name: /of context used/ }) + fireEvent.click(contextTrigger) + const contextPanel = await screen.findByRole('dialog', { name: 'of context used' }) + within(contextPanel).getByText('System prompt') + within(contextPanel).getByText('Tools') + within(contextPanel).getByText('Messages') + // The write/edit turns render a real diff card through the assembled graph // (the keyed FileMutationRow composing ToolRow + DiffBlock), not just the // fixture's raw text. The card is collapsed by default, so expand each edit/ diff --git a/apps/web/tests/math-rendering.e2e.ts b/apps/web/tests/math-rendering.e2e.ts index 2c908bffae..b183fe7df5 100644 --- a/apps/web/tests/math-rendering.e2e.ts +++ b/apps/web/tests/math-rendering.e2e.ts @@ -26,6 +26,7 @@ const DONE = 'MATH_RENDERING_DONE' /** Build a settled assistant reply that exercises every supported math delimiter. */ function mathFixture(): string { const session = Session.create(SessionId('math-rendering-source')) + const eventTimeOrigin = new Date().setHours(12, 0, 0, 0) session.append('turn/start', { turn: 1, }) @@ -76,7 +77,10 @@ function mathFixture(): string { createdAt: 0, cwd: '{{cwd}}', }), - ...session.events.map(event => JSON.stringify(event)), + ...session.events.map(event => JSON.stringify({ + ...event, + time: eventTimeOrigin + event.seq * 1_000, + })), '', ].join('\n') } diff --git a/apps/web/tests/scaffold.ts b/apps/web/tests/scaffold.ts index a96da3745e..52eb7f151d 100644 --- a/apps/web/tests/scaffold.ts +++ b/apps/web/tests/scaffold.ts @@ -476,15 +476,29 @@ export function fixtureUserPrompts(fixtureText: string): string[] { * @param id - the seeded session id (stable for deterministic goldens). * @returns the seeded id. */ -export async function seedSession(scaffold: WebScaffold, fixtureText: string, id: string): Promise { +/** + * Realize a recorded seed fixture against one scaffold: substitute the + * `{{sessionId}}`/`{{cwd}}` placeholders and rewrite the recorded cwd to the + * scaffold's workspace. Idempotent, so a caller may realize early (e.g. to + * price content exactly as the host will fold it) and still pass the result + * through {@link seedSession}. + * @param scaffold - the booted scaffold whose workspace the seed targets. + * @param fixtureText - the committed seed fixture text. + * @param id - the session id the seed is realized for. + * @returns the realized fixture text. + */ +export function realizeSeedFixture(scaffold: WebScaffold, fixtureText: string, id: string): string { const realized = fixtureText .split('{{sessionId}}').join(id) .split('{{cwd}}').join(scaffold.workspaceCwd) const fixtureCwd = (JSON.parse(realized.split('\n', 1)[0]!) as { cwd?: string }).cwd - const rewritten = fixtureCwd === undefined + return fixtureCwd === undefined ? realized : realized.split(fixtureCwd).join(scaffold.workspaceCwd) - const events = parseSessionLog(rewritten) +} + +export async function seedSession(scaffold: WebScaffold, fixtureText: string, id: string): Promise { + const events = parseSessionLog(realizeSeedFixture(scaffold, fixtureText, id)) if (events.length === 0) throw new Error('seed fixture has no events') const last = events[events.length - 1]! // An open final turn would be mutated by resume's crash repair on first @@ -518,8 +532,14 @@ export async function seedSession(scaffold: WebScaffold, fixtureText: string, id } /** - * Normalize an aria snapshot: uuid, cwd, workspace-basename, and duration - * volatility collapse to stable tokens. + * Normalize an aria snapshot: uuid, cwd, workspace-basename, duration, and + * decode-throughput volatility collapse to stable tokens. + * + * Throughput needs a token for the same reason durations do, and no fixture + * can supply one: the figure divides a replayed step's output tokens by the + * wall time the local run took to stream them, so it moves between two runs + * on one machine (measured 69 → 70 tok/s) and swings wildly on a fast replay + * (26333 tok/s for a 3 ms stream). */ function normalizeAria(snapshot: string, workspaceCwd: string): string { // The session heading renders the workspace's basename, not the full @@ -529,14 +549,17 @@ function normalizeAria(snapshot: string, workspaceCwd: string): string { .split(workspaceCwd).join('{{cwd}}') .split(base).join('{{workspace}}') .replace(/[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}/gi, '{{uuid}}') + // The optional space in `\d+m ?\d+s` covers both minute spellings: the + // stats line's compact `2m42s` and the message-chrome template's `2m 42s`. .replace( - /~\d+(?:y(?: \d+mo)?|mo(?: \d+d)?)|\b(?:\d+d(?: \d+h(?: \d+m \d+s)?)?|\d+h \d+m \d+s|\d+m \d+s|\d+(?:\.\d+)?s|\d+(?:\.\d+)?ms)\b/g, + /~\d+(?:y(?: \d+mo)?|mo(?: \d+d)?)|\b(?:\d+d(?: \d+h(?: \d+m \d+s)?)?|\d+h \d+m \d+s|\d+m ?\d+s|\d+(?:\.\d+)?s|\d+(?:\.\d+)?ms)\b/g, duration => duration.startsWith('~') ? duration : '{{duration}}', ) .replace( /约\d+(?:年(?:\d+个月)?|个月(?:\d+天)?)|\d+(?:天(?:\d+小时(?:\d+分\d+秒)?)?|小时\d+分\d+秒|分\d+秒|(?:\.\d+)?秒)/g, duration => duration.startsWith('约') ? duration : '{{duration}}', ) + .replace(/\d+(?:\.\d+)?(?= tok\/s(?!\w))/g, '{{throughput}}') // Message IconActions clocks widen by calendar day/year; collapse every // shape so goldens stay stable across midnight and year boundaries. .replace(/\d{4}年\d{1,2}月\d{1,2}日 \d{2}:\d{2}/g, '{{clock}}') diff --git a/apps/web/tests/seeded-history.e2e.ts b/apps/web/tests/seeded-history.e2e.ts index 128c5303a5..f9ef1055d1 100644 --- a/apps/web/tests/seeded-history.e2e.ts +++ b/apps/web/tests/seeded-history.e2e.ts @@ -16,11 +16,14 @@ import type { Browser, Page } from 'playwright' import { chromium } from 'playwright' import { afterAll, beforeAll, describe, expect, it, onTestFailed } from 'vitest' import { createUserMessage } from '@deepseek-ai/dsh-llm' -import { SessionId } from '@deepseek-ai/dsh-session' +import type { ContentBlock, Message } from '@deepseek-ai/dsh-llm' +import { deriveEventMessage, SessionId } from '@deepseek-ai/dsh-session' +import type { SessionEvent } from '@deepseek-ai/dsh-session' +import type { TokenMeterService } from '@deepseek-ai/dsh-token-meter' import { join } from 'node:path' import { assertFixtureInventory, captureStableAria, compareOrRefreshGolden, fixtureUserPrompts, - launchWebScaffold, recordFixture, seedSession, watchConsole, webSnapshotMode, type WebScaffold, + launchWebScaffold, realizeSeedFixture, recordFixture, seedSession, watchConsole, webSnapshotMode, type WebScaffold, } from './scaffold.ts' import { newEnglishPage, saveFailureShot } from './support.ts' @@ -41,17 +44,22 @@ const PROMPT = 'Use the read tool twice in one assistant message: read a.txt and * deterministic condition before seeding it cold, so the scenario pins the bug * this change fixes — a landed compaction must not erase history the reader * already saw — through the real host and the real browser. - * @param raw - the committed seed fixture text. + * @param raw - the seed fixture text, already realized (placeholder-free) so + * the shadow price below is computed from the exact strings the host folds. + * @param meter - the composed token meter; the appended `compact/summary`'s + * shadow price must be the exact heuristic price of the shadowed nodes, the + * way compact-basic derives it, because the token-meter projections subtract + * it verbatim. * @returns the fixture with a compacted turn appended. */ -function withCompaction(raw: string): string { +function withCompaction(raw: string, meter: TokenMeterService): string { const lines = raw.trimEnd().split('\n') const events = lines.slice(1).map(line => JSON.parse(line) as { type: string seq: number time: number surfaceOp?: unknown - data?: { turn?: unknown } + data?: { turn?: unknown; message?: unknown; content?: unknown; callId?: unknown; isError?: unknown } }) const surfaceSeqs = events .filter(event => event.surfaceOp === 'append' @@ -87,6 +95,31 @@ function withCompaction(raw: string): string { } at({ type: 'turn/start', data: { turn } }) const startSeq = at({ type: 'compact/start', data: { turn } }) + // Load-bearing exactness: the projections subtract this count verbatim, so + // it must equal what the host's fold prices for these nodes. The estimator + // prices message CONTENT only, so a minimal wrapper per storage shape is + // exact — pre-identity rows carry bare `content` (the persistence read path + // upgrades them), a current row carries the full `message` envelope. + const priceRow = (row: (typeof events)[number]): number => { + if (row.data?.message !== undefined) { + const message = deriveEventMessage(row as unknown as SessionEvent) + return message === null ? 0 : meter.estimateMessage(message) + } + const content = row.data?.content as ContentBlock[] + if (row.type === 'tool/result') { + return meter.estimateMessage({ + content: [{ type: 'tool-result', toolCallId: row.data?.callId, content, isError: row.data?.isError === true }], + } as unknown as Message) + } + // An empty-content assistant message derives no transcript entry. + if (row.type === 'assistant/message' && content.length === 0) return 0 + return meter.estimateMessage({ content } as unknown as Message) + } + const shadowedTokenCount = surfaceSeqs.reduce((total, surfaceSeq) => { + const event = events.find(candidate => candidate.seq === surfaceSeq) + if (event === undefined) throw new Error(`seeded-history compaction: shadowed seq ${surfaceSeq} is not in the seed`) + return total + priceRow(event) + }, 0) const summarySeq = at({ type: 'compact/summary', data: { @@ -96,7 +129,7 @@ function withCompaction(raw: string): string { }], shadowedRange: { start: first, end: last }, shadowedSeqs: surfaceSeqs, - shadowedTokenCount: 10_000, + shadowedTokenCount, provider: 'snapshot', model: 'snapshot-compactor', }, @@ -137,7 +170,10 @@ describe('web e2e: seeded history renders through cold resume', () => { if (MODE !== 'record') { const raw = await readFile(SEED, 'utf8') expect(fixtureUserPrompts(raw), 'seed fixture must carry exactly the drive prompt').toEqual([PROMPT]) - await seedSession(scaffold, withCompaction(raw), SEED_ID) + const meter = scaffold.ctx.get('tokenMeter') + if (meter === undefined) throw new Error('seeded-history requires the composed token meter') + const realized = realizeSeedFixture(scaffold, raw, SEED_ID) + await seedSession(scaffold, withCompaction(realized, meter), SEED_ID) } browser = await chromium.launch() page = await newEnglishPage(browser) diff --git a/apps/web/tests/snapshots/bash-abort-row/ui.expected.md b/apps/web/tests/snapshots/bash-abort-row/ui.expected.md index ab680dc776..1b9e6aa339 100644 --- a/apps/web/tests/snapshots/bash-abort-row/ui.expected.md +++ b/apps/web/tests/snapshots/bash-abort-row/ui.expected.md @@ -30,4 +30,4 @@ - text: Select model - img - button "Send message" [disabled] -- text: 1 turns · 1 steps Tool call {{duration}} Cache hit 0% Input 10 tok · Output 10 tok +- text: 1 turns · 1 steps LLM {{duration}} · Tool call {{duration}} TTFT avg {{duration}} · {{throughput}} tok/s Cache hit 0% Input 10 tok · Output 10 tok diff --git a/apps/web/tests/snapshots/code-mode-round/ui.expected.md b/apps/web/tests/snapshots/code-mode-round/ui.expected.md index 3bd030ce88..0c2cf8604c 100644 --- a/apps/web/tests/snapshots/code-mode-round/ui.expected.md +++ b/apps/web/tests/snapshots/code-mode-round/ui.expected.md @@ -36,7 +36,7 @@ - img - button "Branch into a new conversation": - img -- text: {{clock}}Ran for {{duration}} +- text: {{clock}} Ran for {{duration}} TTFT {{duration}} {{throughput}} tok/s - textbox "Message the agent" - button "Commands": - img @@ -44,5 +44,6 @@ - button "Select model, current DeepSeek-V4-Flash": - text: DeepSeek-V4-Flash - img +- button "7% of context used" - button "Send message" [disabled] -- text: 1 turns · 2 steps Tool call {{duration}} Context 7% of 128K Cache hit 52% Input 17.2K tok · Output 252 tok +- text: 1 turns · 2 steps LLM {{duration}} · Tool call {{duration}} TTFT avg {{duration}} · {{throughput}} tok/s Cache hit 52% Input 17.2K tok · Output 252 tok diff --git a/apps/web/tests/snapshots/cordis-tool-round/ui.expected.md b/apps/web/tests/snapshots/cordis-tool-round/ui.expected.md index 57261af55d..33b1d6cd0f 100644 --- a/apps/web/tests/snapshots/cordis-tool-round/ui.expected.md +++ b/apps/web/tests/snapshots/cordis-tool-round/ui.expected.md @@ -51,7 +51,7 @@ - img - button "Branch into a new conversation": - img -- text: {{clock}}Ran for {{duration}} +- text: {{clock}} Ran for {{duration}} TTFT {{duration}} {{throughput}} tok/s - textbox "Message the agent" - button "Commands": - img @@ -59,5 +59,6 @@ - button "Select model, current DeepSeek-V4-Flash": - text: DeepSeek-V4-Flash - img +- button "13% of context used" - button "Send message" [disabled] -- text: 1 turns · 4 steps Tool call {{duration}} Context 13% of 128K Cache hit 77% Input 66.5K tok · Output 312 tok +- text: 1 turns · 4 steps LLM {{duration}} · Tool call {{duration}} TTFT avg {{duration}} · {{throughput}} tok/s Cache hit 77% Input 66.5K tok · Output 312 tok diff --git a/apps/web/tests/snapshots/fresh-round-trip/ui.expected.md b/apps/web/tests/snapshots/fresh-round-trip/ui.expected.md index b259813833..aebc2a45b6 100644 --- a/apps/web/tests/snapshots/fresh-round-trip/ui.expected.md +++ b/apps/web/tests/snapshots/fresh-round-trip/ui.expected.md @@ -31,7 +31,7 @@ - img - button "Branch into a new conversation": - img -- text: {{clock}}Ran for {{duration}} +- text: {{clock}} Ran for {{duration}} TTFT {{duration}} {{throughput}} tok/s - textbox "Message the agent" - button "Commands": - img @@ -39,5 +39,6 @@ - button "Select model, current DeepSeek-V4-Flash": - text: DeepSeek-V4-Flash - img +- button "6% of context used" - button "Send message" [disabled] -- text: 1 turns · 2 steps Tool call {{duration}} Context 6% of 128K Cache hit 99% Input 15.7K tok · Output 111 tok +- text: 1 turns · 2 steps LLM {{duration}} · Tool call {{duration}} TTFT avg {{duration}} · {{throughput}} tok/s Cache hit 99% Input 15.7K tok · Output 111 tok diff --git a/apps/web/tests/snapshots/lifecycle-chrome/reloaded.expected.md b/apps/web/tests/snapshots/lifecycle-chrome/reloaded.expected.md index 1b6e48466a..6b6671ec01 100644 --- a/apps/web/tests/snapshots/lifecycle-chrome/reloaded.expected.md +++ b/apps/web/tests/snapshots/lifecycle-chrome/reloaded.expected.md @@ -23,7 +23,7 @@ - img - button "Branch into a new conversation": - img -- text: {{clock}}Ran for {{duration}} +- text: {{clock}} Ran for {{duration}} TTFT {{duration}} {{throughput}} tok/s - textbox "Message the agent" - button "Commands": - img @@ -31,5 +31,6 @@ - button "Select model, current DeepSeek-V4-Flash": - text: DeepSeek-V4-Flash - img +- button "6% of context used" - button "Send message" [disabled] -- text: 1 turns · 1 steps Context 6% of 128K Cache hit 99% Input 7.8K tok · Output 21 tok +- text: 1 turns · 1 steps LLM {{duration}} TTFT avg {{duration}} · {{throughput}} tok/s Cache hit 99% Input 7.8K tok · Output 21 tok diff --git a/apps/web/tests/snapshots/live-interactions/cancel.expected.md b/apps/web/tests/snapshots/live-interactions/cancel.expected.md index f1c57be871..9735b8acfe 100644 --- a/apps/web/tests/snapshots/live-interactions/cancel.expected.md +++ b/apps/web/tests/snapshots/live-interactions/cancel.expected.md @@ -20,7 +20,7 @@ - img - button "Branch into a new conversation": - img -- text: {{clock}}Ran for {{duration}} +- text: {{clock}} Ran for {{duration}} - textbox "Message the agent" - button "Commands": - img diff --git a/apps/web/tests/snapshots/live-interactions/retry.expected.md b/apps/web/tests/snapshots/live-interactions/retry.expected.md index bcfd99e77b..f127d3e8d1 100644 --- a/apps/web/tests/snapshots/live-interactions/retry.expected.md +++ b/apps/web/tests/snapshots/live-interactions/retry.expected.md @@ -25,7 +25,7 @@ - img - button "Branch into a new conversation": - img -- text: {{clock}}Ran for {{duration}} +- text: {{clock}} Ran for {{duration}} TTFT {{duration}} {{throughput}} tok/s - textbox "Message the agent" - button "Commands": - img @@ -33,5 +33,6 @@ - button "Select model, current DeepSeek-V4-Flash": - text: DeepSeek-V4-Flash - img +- button "6% of context used" - button "Send message" [disabled] -- text: 1 turns · 1 steps Context 6% of 128K Cache hit 99% Input 7.8K tok · Output 79 tok +- text: 1 turns · 1 steps LLM {{duration}} TTFT avg {{duration}} · {{throughput}} tok/s Cache hit 99% Input 7.8K tok · Output 79 tok diff --git a/apps/web/tests/snapshots/markdown-images/ui.expected.md b/apps/web/tests/snapshots/markdown-images/ui.expected.md index ed42dfec84..fbdbff395a 100644 --- a/apps/web/tests/snapshots/markdown-images/ui.expected.md +++ b/apps/web/tests/snapshots/markdown-images/ui.expected.md @@ -19,7 +19,7 @@ - img - button "Branch into a new conversation": - img -- text: {{clock}}Ran for {{duration}} +- text: {{clock}} Ran for {{duration}} - textbox "Message the agent" - button "Commands": - img diff --git a/apps/web/tests/snapshots/math-rendering/ui.expected.md b/apps/web/tests/snapshots/math-rendering/ui.expected.md index f8503988b9..be1bbb7069 100644 --- a/apps/web/tests/snapshots/math-rendering/ui.expected.md +++ b/apps/web/tests/snapshots/math-rendering/ui.expected.md @@ -35,7 +35,7 @@ - img - button "Branch into a new conversation": - img -- text: {{clock}}Ran for {{duration}} +- text: {{clock}} Ran for {{duration}} - textbox "Message the agent" - button "Commands": - img @@ -44,4 +44,4 @@ - text: Select model - img - button "Send message" [disabled] -- text: 1 turns · 1 steps Input 0 tok · Output 0 tok +- text: 1 turns · 1 steps LLM {{duration}} Input 0 tok · Output 0 tok diff --git a/apps/web/tests/snapshots/message-actions/ui.expected.md b/apps/web/tests/snapshots/message-actions/ui.expected.md index 815acaef1d..81c2796e5a 100644 --- a/apps/web/tests/snapshots/message-actions/ui.expected.md +++ b/apps/web/tests/snapshots/message-actions/ui.expected.md @@ -20,7 +20,7 @@ - img - button "Branch into a new conversation" [disabled]: - img -- text: Available only on the last message of a completed turn 7/25 {{clock}}Ran for {{duration}} +- text: Available only on the last message of a completed turn 7/25 {{clock}} Ran for {{duration}} TTFT {{duration}} {{throughput}} tok/s - button "Read a.txt": - img - img @@ -46,7 +46,7 @@ - img - button "Branch into a new conversation": - img -- text: 7/25 {{clock}}Ran for {{duration}} +- text: 7/25 {{clock}} Ran for {{duration}} - textbox "Message the agent" - button "Commands": - img @@ -55,4 +55,4 @@ - text: Select model - img - button "Send message" [disabled] -- text: 2 turns · 3 steps Tool call {{duration}} Cache hit 98% Input 7.8K tok · Output 103 tok +- text: 2 turns · 3 steps LLM {{duration}} · Tool call {{duration}} TTFT avg {{duration}} · {{throughput}} tok/s Cache hit 98% Input 7.8K tok · Output 103 tok diff --git a/apps/web/tests/snapshots/plan-review/approved.expected.md b/apps/web/tests/snapshots/plan-review/approved.expected.md index fe30f4138c..f0c7d718e0 100644 --- a/apps/web/tests/snapshots/plan-review/approved.expected.md +++ b/apps/web/tests/snapshots/plan-review/approved.expected.md @@ -36,7 +36,7 @@ - img - button "Branch into a new conversation": - img -- text: {{clock}}Ran for {{duration}} +- text: {{clock}} Ran for {{duration}} TTFT {{duration}} {{throughput}} tok/s - textbox "Message the agent" - button "Commands": - img @@ -44,5 +44,6 @@ - button "Select model, current DeepSeek-V4-Flash": - text: DeepSeek-V4-Flash - img +- button "4% of context used" - button "Send message" [disabled] -- text: 1 turns · 2 steps Tool call {{duration}} Context 4% of 128K Cache hit 51% Input 10.2K tok · Output 346 tok +- text: 1 turns · 2 steps LLM {{duration}} · Tool call {{duration}} TTFT avg {{duration}} · {{throughput}} tok/s Cache hit 51% Input 10.2K tok · Output 346 tok diff --git a/apps/web/tests/snapshots/question-composer/answered.expected.md b/apps/web/tests/snapshots/question-composer/answered.expected.md index 05c3448ad7..82e0b468c1 100644 --- a/apps/web/tests/snapshots/question-composer/answered.expected.md +++ b/apps/web/tests/snapshots/question-composer/answered.expected.md @@ -31,7 +31,7 @@ - img - button "Branch into a new conversation": - img -- text: {{clock}}Ran for {{duration}} +- text: {{clock}} Ran for {{duration}} TTFT {{duration}} {{throughput}} tok/s - textbox "Message the agent" - button "Commands": - img @@ -39,5 +39,6 @@ - button "Select model, current DeepSeek-V4-Flash": - text: DeepSeek-V4-Flash - img +- button "3% of context used" - button "Send message" [disabled] -- text: 1 turns · 2 steps Tool call {{duration}} Context 3% of 128K Cache hit 95% Input 8.6K tok · Output 180 tok +- text: 1 turns · 2 steps LLM {{duration}} · Tool call {{duration}} TTFT avg {{duration}} · {{throughput}} tok/s Cache hit 95% Input 8.6K tok · Output 180 tok diff --git a/apps/web/tests/snapshots/queue-actions/preserved.expected.md b/apps/web/tests/snapshots/queue-actions/preserved.expected.md index 19d51472d9..e8b65fdea1 100644 --- a/apps/web/tests/snapshots/queue-actions/preserved.expected.md +++ b/apps/web/tests/snapshots/queue-actions/preserved.expected.md @@ -20,7 +20,7 @@ - img - button "Branch into a new conversation": - img -- text: {{clock}}Ran for {{duration}} +- text: {{clock}} Ran for {{duration}} - button "2 queued messages" [expanded] - list: - listitem: diff --git a/apps/web/tests/snapshots/seeded-history/command-row.expected.md b/apps/web/tests/snapshots/seeded-history/command-row.expected.md index 4783226424..467a4364b8 100644 --- a/apps/web/tests/snapshots/seeded-history/command-row.expected.md +++ b/apps/web/tests/snapshots/seeded-history/command-row.expected.md @@ -33,7 +33,7 @@ - img - button "Branch into a new conversation": - img -- text: 7/25 {{clock}}Ran for {{duration}} +- text: 7/25 {{clock}} Ran for {{duration}} TTFT {{duration}} {{throughput}} tok/s - button "Context compacted View compaction summary": - img - text: Context compacted View compaction summary @@ -51,4 +51,4 @@ - text: Select model - img - button "Send message" [disabled] -- text: 1 turns · 2 steps Tool call {{duration}} Cache hit 98% Input 15.8K tok · Output 135 tok +- text: 1 turns · 2 steps LLM {{duration}} · Tool call {{duration}} TTFT avg {{duration}} · {{throughput}} tok/s Cache hit 98% Input 15.8K tok · Output 135 tok diff --git a/apps/web/tests/snapshots/seeded-history/ui.expected.md b/apps/web/tests/snapshots/seeded-history/ui.expected.md index 65587f37b5..55fcb89ec8 100644 --- a/apps/web/tests/snapshots/seeded-history/ui.expected.md +++ b/apps/web/tests/snapshots/seeded-history/ui.expected.md @@ -33,7 +33,7 @@ - img - button "Branch into a new conversation": - img -- text: 7/25 {{clock}}Ran for {{duration}} +- text: 7/25 {{clock}} Ran for {{duration}} TTFT {{duration}} {{throughput}} tok/s - button "Context compacted View compaction summary": - img - text: Context compacted View compaction summary @@ -49,4 +49,4 @@ - text: Select model - img - button "Send message" [disabled] -- text: 1 turns · 2 steps Tool call {{duration}} Cache hit 98% Input 15.8K tok · Output 135 tok +- text: 1 turns · 2 steps LLM {{duration}} · Tool call {{duration}} TTFT avg {{duration}} · {{throughput}} tok/s Cache hit 98% Input 15.8K tok · Output 135 tok diff --git a/apps/web/tests/snapshots/steering/settled.expected.md b/apps/web/tests/snapshots/steering/settled.expected.md index c20cab773a..77385c6333 100644 --- a/apps/web/tests/snapshots/steering/settled.expected.md +++ b/apps/web/tests/snapshots/steering/settled.expected.md @@ -37,7 +37,7 @@ - img - button "Branch into a new conversation": - img -- text: {{clock}}Ran for {{duration}} +- text: {{clock}} Ran for {{duration}} TTFT {{duration}} {{throughput}} tok/s - textbox "Message the agent" - button "Commands": - img @@ -45,5 +45,6 @@ - button "Select model, current DeepSeek-V4-Flash": - text: DeepSeek-V4-Flash - img +- button "6% of context used" - button "Send message" [disabled] -- text: 1 turns · 2 steps Tool call {{duration}} Context 6% of 128K Cache hit 98% Input 15.8K tok · Output 156 tok +- text: 1 turns · 2 steps LLM {{duration}} · Tool call {{duration}} TTFT avg {{duration}} · {{throughput}} tok/s Cache hit 98% Input 15.8K tok · Output 156 tok diff --git a/apps/web/tests/snapshots/subagent-conversation/ui.expected.md b/apps/web/tests/snapshots/subagent-conversation/ui.expected.md index 401061a3b6..a01eea56d8 100644 --- a/apps/web/tests/snapshots/subagent-conversation/ui.expected.md +++ b/apps/web/tests/snapshots/subagent-conversation/ui.expected.md @@ -28,7 +28,7 @@ - img - button "Branch into a new conversation": - img -- text: {{clock}}Ran for {{duration}} Now give the same explanation to a human reader. {{clock}} +- text: {{clock}} Ran for {{duration}} TTFT {{duration}} {{throughput}} tok/s Now give the same explanation to a human reader. {{clock}} - button "Copy": - img - button "Branch into a new conversation" [disabled]: @@ -43,10 +43,11 @@ - img - button "Branch into a new conversation": - img -- text: {{clock}}Ran for {{duration}} +- text: {{clock}} Ran for {{duration}} TTFT {{duration}} {{throughput}} tok/s - textbox "Message the agent" - button "Commands": - img - 'button "Access mode, current: Workspace Write"': Workspace Write +- button "6% of context used" - button "Send message" [disabled] -- text: 2 turns · 2 steps Context 6% of 128K Cache hit 99% Input 15.6K tok · Output 158 tok +- text: 2 turns · 2 steps LLM {{duration}} TTFT avg {{duration}} · {{throughput}} tok/s Cache hit 99% Input 15.6K tok · Output 158 tok diff --git a/apps/web/tests/snapshots/web-search-round/ui.expected.md b/apps/web/tests/snapshots/web-search-round/ui.expected.md index 2b355156e8..1e2dcf9eca 100644 --- a/apps/web/tests/snapshots/web-search-round/ui.expected.md +++ b/apps/web/tests/snapshots/web-search-round/ui.expected.md @@ -23,7 +23,7 @@ - img - button "Branch into a new conversation": - img -- text: {{clock}}Ran for {{duration}} +- text: {{clock}} Ran for {{duration}} TTFT {{duration}} {{throughput}} tok/s - textbox "Message the agent" - button "Commands": - img @@ -31,5 +31,6 @@ - button "Select model, current DeepSeek-V4-Flash": - text: DeepSeek-V4-Flash - img +- button "0% of context used" - button "Send message" [disabled] -- text: 1 turns · 2 steps Tool call {{duration}} Context 0% of 128K Cache hit 0% Input 22 tok · Output 7 tok +- text: 1 turns · 2 steps LLM {{duration}} · Tool call {{duration}} TTFT avg {{duration}} · {{throughput}} tok/s Cache hit 0% Input 22 tok · Output 7 tok diff --git a/docs/config-catalog.md b/docs/config-catalog.md index a261410feb..ffa3d5bdd4 100644 --- a/docs/config-catalog.md +++ b/docs/config-catalog.md @@ -405,6 +405,8 @@ Source: [`packages/compact/compact-basic/src/types.ts:38`](../packages/compact/c ## `@deepseek-ai/dsh-compact-tool-result-prune` +Requires: `tokenMeter` + ```ts config-catalog /** Character-budget policy for deterministic tool-result pruning. */ export interface ToolResultPruneConfig { diff --git a/docs/cordis-catalog/services.md b/docs/cordis-catalog/services.md index 5913c244b8..ed615ed377 100644 --- a/docs/cordis-catalog/services.md +++ b/docs/cordis-catalog/services.md @@ -1707,7 +1707,7 @@ fork(source: SessionForkSource, boundary?: number, childSessionId?: SessionId): Types: [CreateSessionOptions](../core-data-structures/persistence.md) · [PrepareSessionOptions](../core-data-structures/persistence.md) · [Session](../core-data-structures/session.md) · [SessionId](../core-data-structures/core.md) -Source: [`packages/core/session/src/index.ts:837`](../../packages/core/session/src/index.ts) +Source: [`packages/core/session/src/index.ts:800`](../../packages/core/session/src/index.ts) ## `ctx.sessionTitle` — `SessionTitleService` @@ -2339,7 +2339,8 @@ Replay owner for one service-wide estimator and isolated per-session folds. measure(session: Session, requestHeader?: EpochHeader): TokenMeasurement /** - * Heuristically price one model-visible message. + * Heuristically price one model-visible message (instance face of the pure + * `estimateMessage` export from `estimate.ts`). * @param message - message to price without mutation. * @returns content and role-framing tokens under the fixed service heuristic. */ @@ -2348,7 +2349,7 @@ estimateMessage(message: Message): number Types: [EpochHeader](../core-data-structures/session.md) · [Message](../core-data-structures/core.md) · [Session](../core-data-structures/session.md) · [TokenMeasurement](../core-data-structures/token-meter.md) -Source: [`packages/llm/token-meter/src/index.ts:85`](../../packages/llm/token-meter/src/index.ts) +Source: [`packages/llm/token-meter/src/index.ts:74`](../../packages/llm/token-meter/src/index.ts) ## `ctx.toolResultPrune` — `ToolResultPruneService` @@ -2374,7 +2375,10 @@ pruneContent(blocks: readonly ContentBlock[]): ContentBlock[] | null /** * Prune every over-budget tool result from one stable current-surface snapshot. * Each replacement preserves the complete event data except for `content`, - * and points at the shadowed node for durable provenance and replay. + * points at the shadowed node for durable provenance and replay, and is + * immediately preceded by a `compact/prune` shadow-price event pricing the + * shadowed node through the injected token meter, so pure consumers can + * subtract it without per-node state. * @param session - session whose current surface is rewritten. * @returns landed replacements and aggregate Unicode-code-point savings. * @throws when the session rejects a replacement; replacements committed @@ -2385,7 +2389,7 @@ pruneSession(session: Session): PruneResult Types: [ContentBlock](../core-data-structures/core.md) · [PruneResult](../core-data-structures/compaction.md) · [Session](../core-data-structures/session.md) -Source: [`packages/compact/compact-tool-result-prune/src/index.ts:40`](../../packages/compact/compact-tool-result-prune/src/index.ts) +Source: [`packages/compact/compact-tool-result-prune/src/index.ts:44`](../../packages/compact/compact-tool-result-prune/src/index.ts) ## `ctx.tools` — `ToolRegistry` diff --git a/docs/core-data-structures/session.i18n.yaml b/docs/core-data-structures/session.i18n.yaml index 565f7275f6..87f705345c 100644 --- a/docs/core-data-structures/session.i18n.yaml +++ b/docs/core-data-structures/session.i18n.yaml @@ -2,5 +2,5 @@ # side as of the last confirmed-consistent state. Both languages carry equal authority; # after editing either side, bring the other along and re-record with: # pnpm run verify-translation-pairing --write docs/core-data-structures/session.md -session.md: 30f9d7a92f36b0649ec6d61bb3e69a80b125cc73 -session.zh.md: 80762f097bad5f6ab81f3872df5c8b715109241f +session.md: fac201b581e395865fd46d51bca1350cbbc46e8e +session.zh.md: 53dc9d11de895deec72aaf5ea81c70ba87c9c6bd diff --git a/docs/core-data-structures/session.md b/docs/core-data-structures/session.md index 30f9d7a92f..fac201b581 100644 --- a/docs/core-data-structures/session.md +++ b/docs/core-data-structures/session.md @@ -494,15 +494,8 @@ declare class Session { */ deriveMessages(): Message[]; /** - * Project a single event into the LLM message it derives to, or null when - * it produces none — a non-surface event (chunk, boundary, log-only record) - * or an empty-content assistant/message (which exists only to host usage). - * The per-node pure function {@link deriveMessages} folds over the surface; - * an external reconstructor (or the dev invariant) folds the same function - * over a log prefix's surface to rebuild the exact messages any request was - * built from (the reconstructability Agent Note). The returned message is - * the already frozen message nested in the event wrapper and shared by - * delivery, durable history, and model requests. + * Instance face of the pure per-node `deriveEventMessage` export from + * `surface.ts`. * @param event - the event to project. * @returns the derived message, or null when the event produces none. */ diff --git a/docs/core-data-structures/session.zh.md b/docs/core-data-structures/session.zh.md index 80762f097b..53dc9d11de 100644 --- a/docs/core-data-structures/session.zh.md +++ b/docs/core-data-structures/session.zh.md @@ -496,15 +496,8 @@ declare class Session { */ deriveMessages(): Message[]; /** - * Project a single event into the LLM message it derives to, or null when - * it produces none — a non-surface event (chunk, boundary, log-only record) - * or an empty-content assistant/message (which exists only to host usage). - * The per-node pure function {@link deriveMessages} folds over the surface; - * an external reconstructor (or the dev invariant) folds the same function - * over a log prefix's surface to rebuild the exact messages any request was - * built from (the reconstructability Agent Note). The returned message is - * the already frozen message nested in the event wrapper and shared by - * delivery, durable history, and model requests. + * Instance face of the pure per-node `deriveEventMessage` export from + * `surface.ts`. * @param event - the event to project. * @returns the derived message, or null when the event produces none. */ diff --git a/docs/module-graph.md b/docs/module-graph.md index 8ead055d4b..0e2e7e0c37 100644 --- a/docs/module-graph.md +++ b/docs/module-graph.md @@ -428,9 +428,6 @@ flowchart TD pkg_compact --> pkg_invariants pkg_compact --> pkg_llm pkg_compact --> pkg_session - pkg_compact_tool_result_prune --> pkg_invariants - pkg_compact_tool_result_prune --> pkg_llm - pkg_compact_tool_result_prune --> pkg_session pkg_web_fetch_local --> pkg_invariants pkg_web_fetch_local --> pkg_timeout pkg_web_fetch_local --> pkg_web @@ -490,6 +487,7 @@ flowchart TD pkg_llm_retry --> pkg_llm pkg_llm_retry --> pkg_session pkg_llm_retry --> pkg_timeout + pkg_token_meter --> pkg_compact pkg_token_meter --> pkg_invariants pkg_token_meter --> pkg_llm pkg_token_meter --> pkg_session @@ -626,13 +624,11 @@ flowchart TD pkg_command_compact --> pkg_commands pkg_command_compact --> pkg_compact pkg_command_compact --> pkg_invariants - pkg_compact_basic --> pkg_agent - pkg_compact_basic --> pkg_compact - pkg_compact_basic --> pkg_compact_tool_result_prune - pkg_compact_basic --> pkg_invariants - pkg_compact_basic --> pkg_llm - pkg_compact_basic --> pkg_session - pkg_compact_basic --> pkg_token_meter + pkg_compact_tool_result_prune --> pkg_compact + pkg_compact_tool_result_prune --> pkg_invariants + pkg_compact_tool_result_prune --> pkg_llm + pkg_compact_tool_result_prune --> pkg_session + pkg_compact_tool_result_prune --> pkg_token_meter pkg_session_query --> pkg_brand pkg_session_query --> pkg_invariants pkg_session_query --> pkg_llm @@ -728,6 +724,13 @@ flowchart TD pkg_tool_skill --> pkg_llm pkg_tool_skill --> pkg_skill pkg_tool_skill --> pkg_tools + pkg_compact_basic --> pkg_agent + pkg_compact_basic --> pkg_compact + pkg_compact_basic --> pkg_compact_tool_result_prune + pkg_compact_basic --> pkg_invariants + pkg_compact_basic --> pkg_llm + pkg_compact_basic --> pkg_session + pkg_compact_basic --> pkg_token_meter pkg_subagent --> pkg_agent pkg_subagent --> pkg_brand pkg_subagent --> pkg_invariants @@ -1138,7 +1141,6 @@ flowchart TD | [`bash`](../packages/bash/bash) | `bash` | [`invariants`](../packages/support/invariants), [`sandbox`](../packages/sandbox/sandbox), [`subprocess`](../packages/subprocess/subprocess) | | [`fs`](../packages/fs/fs) | `fs` | [`brand`](../packages/util/brand), [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`sandbox`](../packages/sandbox/sandbox) | | [`compact`](../packages/compact/compact) | `compact` | [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`session`](../packages/core/session) | -| [`compact-tool-result-prune`](../packages/compact/compact-tool-result-prune) | `compact` | [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`session`](../packages/core/session) | | [`web-fetch-local`](../packages/web/web-fetch-local) | `web` | [`invariants`](../packages/support/invariants), [`timeout`](../packages/util/timeout), [`web`](../packages/web/web) | | [`web-search-exa`](../packages/web/web-search-exa) | `web` | [`invariants`](../packages/support/invariants), [`web`](../packages/web/web) | | [`web-search-perplexity`](../packages/web/web-search-perplexity) | `web` | [`invariants`](../packages/support/invariants), [`web`](../packages/web/web) | @@ -1155,7 +1157,7 @@ flowchart TD | [`sandbox-local`](../packages/sandbox/sandbox-local) | `sandbox` | [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`sandbox`](../packages/sandbox/sandbox) | | [`session-projection`](../packages/session-projection/session-projection) | `session-projection` | [`invariants`](../packages/support/invariants), [`session`](../packages/core/session) | | [`llm-retry`](../packages/llm/llm-retry) | `llm` | [`agent`](../packages/core/agent), [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`session`](../packages/core/session), [`timeout`](../packages/util/timeout) | -| [`token-meter`](../packages/llm/token-meter) | `llm` | [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`session`](../packages/core/session), [`session-projection`](../packages/session-projection/session-projection) | +| [`token-meter`](../packages/llm/token-meter) | `llm` | [`compact`](../packages/compact/compact), [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`session`](../packages/core/session), [`session-projection`](../packages/session-projection/session-projection) | | [`goal`](../packages/goal/goal) | `goal` | [`agent`](../packages/core/agent), [`brand`](../packages/util/brand), [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`scope`](../packages/core/scope), [`session`](../packages/core/session), [`session-projection`](../packages/session-projection/session-projection) | | [`bash-local`](../packages/bash/bash-local) | `bash` | [`bash`](../packages/bash/bash), [`invariants`](../packages/support/invariants), [`subprocess`](../packages/subprocess/subprocess), [`timeout`](../packages/util/timeout) | | [`pwsh-local`](../packages/bash/pwsh-local) | `bash` | [`bash`](../packages/bash/bash), [`invariants`](../packages/support/invariants), [`subprocess`](../packages/subprocess/subprocess), [`timeout`](../packages/util/timeout) | @@ -1188,7 +1190,7 @@ flowchart TD | [`bash-sandbox`](../packages/bash/bash-sandbox) | `bash` | [`bash`](../packages/bash/bash), [`bash-local`](../packages/bash/bash-local), [`invariants`](../packages/support/invariants), [`sandbox`](../packages/sandbox/sandbox), [`sandbox-policy`](../packages/sandbox/sandbox-policy) | | [`fs-sandbox`](../packages/fs/fs-sandbox) | `fs` | [`fs`](../packages/fs/fs), [`fs-local`](../packages/fs/fs-local), [`invariants`](../packages/support/invariants), [`sandbox`](../packages/sandbox/sandbox), [`sandbox-policy`](../packages/sandbox/sandbox-policy) | | [`command-compact`](../packages/compact/command-compact) | `compact` | [`commands`](../packages/ui/commands), [`compact`](../packages/compact/compact), [`invariants`](../packages/support/invariants) | -| [`compact-basic`](../packages/compact/compact-basic) | `compact` | [`agent`](../packages/core/agent), [`compact`](../packages/compact/compact), [`compact-tool-result-prune`](../packages/compact/compact-tool-result-prune), [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`session`](../packages/core/session), [`token-meter`](../packages/llm/token-meter) | +| [`compact-tool-result-prune`](../packages/compact/compact-tool-result-prune) | `compact` | [`compact`](../packages/compact/compact), [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`session`](../packages/core/session), [`token-meter`](../packages/llm/token-meter) | | [`session-query`](../packages/session-query/session-query) | `session-query` | [`brand`](../packages/util/brand), [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`session`](../packages/core/session), [`session-persistence`](../packages/session-persistence/session-persistence), [`session-title`](../packages/session-title/session-title) | | [`session-title-llm`](../packages/session-title/session-title-llm) | `session-title` | [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`session`](../packages/core/session), [`session-title`](../packages/session-title/session-title), [`timeout`](../packages/util/timeout) | | [`acp`](../packages/acp/acp) | `acp` | [`agent`](../packages/core/agent), [`invariants`](../packages/support/invariants), [`session`](../packages/core/session), [`user-approval`](../packages/ui/user-approval) | @@ -1204,6 +1206,7 @@ flowchart TD | [`tool-fs-search`](../packages/fs/tool-fs-search) | `fs` | [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`retention`](../packages/util/retention), [`session`](../packages/core/session), [`spill`](../packages/spill/spill), [`subprocess`](../packages/subprocess/subprocess), [`system-prompt`](../packages/core/system-prompt), [`tools`](../packages/core/tools) | | [`tool-str-replace-editor`](../packages/fs/tool-str-replace-editor) | `fs` | [`fs`](../packages/fs/fs), [`invariants`](../packages/support/invariants), [`sandbox`](../packages/sandbox/sandbox), [`sandbox-policy`](../packages/sandbox/sandbox-policy), [`tools`](../packages/core/tools) | | [`tool-skill`](../packages/skill/tool-skill) | `skill` | [`agent`](../packages/core/agent), [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`skill`](../packages/skill/skill), [`tools`](../packages/core/tools) | +| [`compact-basic`](../packages/compact/compact-basic) | `compact` | [`agent`](../packages/core/agent), [`compact`](../packages/compact/compact), [`compact-tool-result-prune`](../packages/compact/compact-tool-result-prune), [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`session`](../packages/core/session), [`token-meter`](../packages/llm/token-meter) | | [`subagent`](../packages/subagent/subagent) | `subagent` | [`agent`](../packages/core/agent), [`brand`](../packages/util/brand), [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`scope`](../packages/core/scope), [`session`](../packages/core/session), [`session-persistence`](../packages/session-persistence/session-persistence), [`session-projection`](../packages/session-projection/session-projection), [`session-query`](../packages/session-query/session-query), [`tasks`](../packages/tasks/tasks), [`tools`](../packages/core/tools) | | [`tool-web`](../packages/web/tool-web) | `web` | [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`system-prompt`](../packages/core/system-prompt), [`tools`](../packages/core/tools), [`web`](../packages/web/web) | | [`spill-policy`](../packages/spill/spill-policy) | `spill` | [`invariants`](../packages/support/invariants), [`llm`](../packages/llm/llm), [`retention`](../packages/util/retention), [`session`](../packages/core/session), [`spill`](../packages/spill/spill), [`tools`](../packages/core/tools) | diff --git a/docs/persistence-catalog.md b/docs/persistence-catalog.md index caff1e7685..f033fde18f 100644 --- a/docs/persistence-catalog.md +++ b/docs/persistence-catalog.md @@ -236,7 +236,31 @@ Source: [`packages/ui/commands/src/index.ts:132`](../packages/ui/commands/src/in 'compact/end': { turn: number | null; error?: string } ``` -Source: [`packages/compact/compact/src/types.ts:51`](../packages/compact/compact/src/types.ts) +Source: [`packages/compact/compact/src/types.ts:54`](../packages/compact/compact/src/types.ts) + +#### `compact/prune` — log-only + +```ts persistence-catalog +/** + * Shadow price of one model-free prune replacement — log-only, no + * surfaceOp. The shared shadow-price protocol: a surface `replace` event + * is priced by the metering event immediately before it (`compact/summary` + * for a summarizing compaction, this event for a prune), which states the + * heuristic token price of the exact replaced range so a pure consumer + * can subtract it without retaining per-node prices. The replacement MUST + * be appended synchronously right after this event. + */ +'compact/prune': { + /** The replaced range's first and last surface-node seqs (a surface-position span, like {@link CompactionResult.shadowedRange}). */ + shadowedRange: { start: number; end: number } + /** The seqs of all shadowed surface nodes, in surface order. */ + shadowedSeqs: number[] + /** Heuristic price of the shadowed content under the token-meter's fixed estimator. */ + shadowedTokenCount: number +} +``` + +Source: [`packages/compact/compact/src/types.ts:64`](../packages/compact/compact/src/types.ts) #### `compact/start` — log-only @@ -257,8 +281,11 @@ Source: [`packages/compact/compact/src/types.ts:19`](../packages/compact/compact /** * Provenance record of a completed summarization — log-only, no surfaceOp. * The summary content is in `data.summary`; the actual surface replacement - * is performed by a subsequent `user/message` event that shadows the - * compacted range. + * is performed by the immediately following `user/message` event that + * shadows the compacted range. That adjacency is contractual — the + * shadowed pricing fields are the replacement's shadow price, so a + * consumer may pair a replacement with the metering event directly + * before it (`compact/prune` documents the shared protocol). */ 'compact/summary': { summary: ContentBlock[] @@ -285,7 +312,7 @@ Source: [`packages/compact/compact/src/types.ts:19`](../packages/compact/compact Types: [ContentBlock](core-data-structures/core.md) · [TokenUsage](core-data-structures/llm-streaming.md) -Source: [`packages/compact/compact/src/types.ts:26`](../packages/compact/compact/src/types.ts) +Source: [`packages/compact/compact/src/types.ts:29`](../packages/compact/compact/src/types.ts) ### `goal/*` diff --git a/packages/client/connection/src/client/fixture.ts b/packages/client/connection/src/client/fixture.ts index 45ae45928c..be81b54b17 100644 --- a/packages/client/connection/src/client/fixture.ts +++ b/packages/client/connection/src/client/fixture.ts @@ -27,7 +27,7 @@ import type { // Type-only: the brand constructor is host-side; the fixture casts at its // wire-fabrication boundary (the schema layer's one-cast-point posture). import type { CommandId } from '@deepseek-ai/dsh-commands/brand' -import { foldSurface } from '@deepseek-ai/dsh-session/surface' +import { deriveEventMessage, foldSurface } from '@deepseek-ai/dsh-session/surface' import type { ApiProxy, ClientRequest, ClientResponse, HistoryEntry, HostFrame, MuxFrame, RpcReceipt, ModelProviderGroup, ModelTarget, RpcRequest, RpcResponse, RpcResult, ServerRequest, ServerResponse, SessionSummary, @@ -358,6 +358,12 @@ function buildAlphaLog(): SessionEvent[] { events.push({ seq, time: (time += 800), ...authored }) return seq } + // This resident history represents completed model requests, so retain the + // route capacity that accompanied them just as the live prompt path does. + push({ + type: 'request/context', + data: { provider: 'deepseek-official', model: 'deepseek-v4-flash', contextWindow: 128_000 }, + }) for (let turn = 0; turn < 60; turn++) { push({ type: 'turn/start', data: { turn } }) const userSeq = push({ @@ -819,6 +825,62 @@ interface FixtureRequestContext { contextWindow?: number } +interface FixtureContextBreakdownProjection { + systemTokens: number + toolsTokens: number + messageTokens: number +} + +/** Fixed token-meter heuristic constants mirrored by this client-only fixture. */ +const CHARS_PER_TOKEN = 4 +const BLOCK_OVERHEAD = 4 +const ROLE_OVERHEAD = 4 + +/** Price fixture content with token-meter's fixed-density heuristic. */ +function estimateFixtureContent(blocks: readonly ContentBlock[]): number { + const densityPrice = (value: string): number => Math.ceil(value.length / CHARS_PER_TOKEN) + return blocks.reduce((tokens, block) => { + if (block.type === 'text' || block.type === 'reasoning') { + return tokens + densityPrice(block.text) + BLOCK_OVERHEAD + } + if (block.type === 'tool-call') { + return tokens + densityPrice(block.name) + densityPrice(block.arguments) + BLOCK_OVERHEAD + } + // ContentBlockMap is merge-extensible: this client graph sees only the + // base four members, but fixture turns do carry extended blocks at + // runtime, so the structural JSON fallback below is live code. + // oxlint-disable-next-line typescript/no-unnecessary-condition -- the type collapses without the out-of-graph merges (see above). + if (block.type === 'tool-result') { + return tokens + estimateFixtureContent(block.content) + BLOCK_OVERHEAD + } + return tokens + densityPrice(JSON.stringify(block)) + BLOCK_OVERHEAD + }, 0) +} + +/** Fixture parallel of token-meter's heuristic context-composition projection. */ +function contextBreakdownOf(log: readonly SessionEvent[]): FixtureContextBreakdownProjection { + const headerEvent = log.findLast(event => event.type === 'request/header') + const header = headerEvent === undefined + ? undefined + : headerEvent.data.header + let messageTokens = 0 + for (const seq of foldSurface(log).nodes) { + const event = log[seq] + if (event === undefined) continue + const message = deriveEventMessage(event) + if (message !== null) messageTokens += estimateFixtureContent(message.content) + ROLE_OVERHEAD + } + return { + systemTokens: header?.system === undefined + ? 0 + : Math.ceil(header.system.length / CHARS_PER_TOKEN) + ROLE_OVERHEAD, + toolsTokens: header?.tools === undefined || header.tools.length === 0 + ? 0 + : Math.ceil(JSON.stringify(header.tools).length / CHARS_PER_TOKEN) + BLOCK_OVERHEAD, + messageTokens, + } +} + /** Latest log-only route context, or undefined before any request ran. */ function lastRequestContext( log: readonly SessionEvent[], @@ -832,7 +894,11 @@ function lastRequestContext( /** * Fixture parallel of token-meter's request-pressure projection: the last * provider-reported prompt size paired with the last recorded capacity. The - * two need not come from one request — see the token-meter README. + * two need not come from one request — see the token-meter README. The host's + * `projectedTokens` is deliberately absent: reproducing it would mean + * reimplementing the estimator client-side, and every consumer falls back to + * the bare sample, so a fixture-driven view simply lags a compaction the way + * the projection did before that field existed. */ function contextPressureOf( log: readonly SessionEvent[], @@ -870,28 +936,44 @@ function projectionValuesOf(log: readonly SessionEvent[]): Record[] { const type = (event as { type: string }).type + const frames: Extract[] = [] // One usage sample advances both token-meter units. if (usageSampleOf(event) !== undefined) { - return [ + frames.push( { type: 'session/projection', sessionId: id, key: 'tokenUsage', value: tokenUsageOf(log), seq: event.seq }, { type: 'session/projection', sessionId: id, key: 'contextPressure', value: contextPressureOf(log), seq: event.seq }, - ] + ) } if (type === 'request/context') { - return [{ + frames.push({ type: 'session/projection', sessionId: id, key: 'contextPressure', value: contextPressureOf(log), seq: event.seq, - }] + }) } + if (type === 'request/header' + || type === 'user/message' + || type === 'assistant/message' + || type === 'tool/result') { + frames.push({ + type: 'session/projection', + sessionId: id, + key: 'contextBreakdown', + value: contextBreakdownOf(log), + seq: event.seq, + }) + } + if (frames.length > 0) return frames if (type === 'session/title') { const values = projectionValuesOf(log) /* v8 ignore next -- the advancing title event is in the log, so the key is present. */ diff --git a/packages/client/connection/tests/fixture.spec.ts b/packages/client/connection/tests/fixture.spec.ts index ad95f5bb17..3bdd0a21ea 100644 --- a/packages/client/connection/tests/fixture.spec.ts +++ b/packages/client/connection/tests/fixture.spec.ts @@ -159,6 +159,11 @@ describe('createFixtureApi', () => { }, // No request ran, so neither pressure nor capacity is known yet. contextPressure: {}, + contextBreakdown: { + systemTokens: 0, + toolsTokens: 0, + messageTokens: 0, + }, } }, }) }) @@ -304,6 +309,10 @@ describe('createFixtureApi', () => { frame.type === 'session/projection' && frame.key === 'contextPressure' && (frame.value as { contextWindow?: number }).contextWindow === 128_000)).toBe(true) + expect(frames.some(frame => + frame.type === 'session/projection' + && frame.key === 'contextBreakdown' + && (frame.value as { messageTokens?: number }).messageTokens! > 0)).toBe(true) const finalize = frames.find((f): f is Extract => f.type === 'session/event' && f.event.type === 'assistant/message') expect(JSON.stringify(finalize?.event.data)).toContain('（已中断）') // Idle cancel: no replay in flight, must not explode; running flips false. @@ -335,7 +344,7 @@ describe('createFixtureApi', () => { const envelopes: RpcRequest[] = [] for await (const envelope of api.events.mux(req({}), abort.signal)) { envelopes.push(envelope) - if (envelopes.length >= 10) abort.abort() + if (envelopes.length >= 11) abort.abort() } return envelopes } @@ -351,10 +360,15 @@ describe('createFixtureApi', () => { expect(first[5]?.payload).toMatchObject({ type: 'session/projection', sessionId: 'fx-alpha', key: 'goal', value: null }) expect(first[6]?.payload).toMatchObject({ type: 'session/projection', sessionId: 'fx-alpha', key: 'tokenUsage' }) expect(first[7]?.payload).toMatchObject({ type: 'session/projection', sessionId: 'fx-alpha', key: 'contextPressure' }) - expect(first[8]?.payload).toMatchObject({ type: 'approval/requested', toolName: 'dangerous_tool' }) - expect(second[8]?.rpcId).toBe(first[8]?.rpcId) // stable rpcId across replays (host replay semantics) - expect(first[9]?.payload).toMatchObject({ type: 'question/requested', sessionId: 'fx-alpha' }) - expect(second[9]?.rpcId).toBe(first[9]?.rpcId) + expect(first[8]?.payload).toMatchObject({ + type: 'session/projection', sessionId: 'fx-alpha', key: 'contextBreakdown', + value: { systemTokens: 0, toolsTokens: 0 }, + }) + expect((first[8]?.payload as { value: { messageTokens: number } }).value.messageTokens).toBeGreaterThan(0) + expect(first[9]?.payload).toMatchObject({ type: 'approval/requested', toolName: 'dangerous_tool' }) + expect(second[9]?.rpcId).toBe(first[9]?.rpcId) // stable rpcId across replays (host replay semantics) + expect(first[10]?.payload).toMatchObject({ type: 'question/requested', sessionId: 'fx-alpha' }) + expect(second[10]?.rpcId).toBe(first[10]?.rpcId) }) it('steer with no replay in flight falls through to a fresh queued turn; non-text blocks stringify empty', async () => { diff --git a/packages/client/runtime/src/client/session-history/history-fold.ts b/packages/client/runtime/src/client/session-history/history-fold.ts index 56c5f31c26..d792fd2b76 100644 --- a/packages/client/runtime/src/client/session-history/history-fold.ts +++ b/packages/client/runtime/src/client/session-history/history-fold.ts @@ -18,6 +18,8 @@ import type { } from '../sessions/conversation-context.ts' import type { ConversationPromptSnapshot } from '../sessions/request-inspection.ts' import { PartialAccumulator } from '../sessions/partial.ts' +import type { AssistantStepMetadata } from '../sessions/assistant-timing.ts' +import { indexAssistantStepTiming, settledAssistantTiming } from '../sessions/assistant-timing.ts' interface CallIndexEntry { name: string @@ -32,11 +34,6 @@ interface FoldedContext { originSeq?: number } -interface AssistantStepMetadata { - stepStartTime: number | null - firstTokenTime: number | null -} - /** Immutable conversation projections derived only from the history source. */ export interface ConversationHistoryProjection { eventNodes: readonly ConversationNode[] @@ -47,10 +44,6 @@ export interface ConversationHistoryProjection { codeDispatches: ReadonlyMap } -function assistantStepKey(turn: number, step: number): string { - return `${turn}\u0000${step}` -} - function replacementCrossesWindowHead(event: SessionEvent, baseSeq: number): boolean { if (!isSurfaceEvent(event) || event.surfaceOp === 'append') return false return event.surfaceOp.start < baseSeq || event.surfaceOp.end < baseSeq @@ -66,21 +59,7 @@ function contextOriginKind(event: SessionEvent | undefined): ConversationContext return 'rewrite' } -function isTokenDelta(chunk: SessionEvent<'assistant/chunk'>['data']['chunk']): boolean { - switch (chunk.type) { - case 'text-delta': - case 'reasoning-delta': - return chunk.text !== '' - case 'tool-call-delta': - return chunk.argumentsDelta !== '' || chunk.name !== undefined - default: - return false - } -} - -function foldContexts( - events: readonly SessionEvent[], -): readonly FoldedContext[] { +function foldContexts(events: readonly SessionEvent[]): readonly FoldedContext[] { const replay: SessionEvent[] = [] const originalSeqs: number[] = [] const rebasedSeqByOriginal = new Map() @@ -398,6 +377,7 @@ export function projectConversationHistory( contextGeneration++ if (activePrompt !== undefined) promptsByContext.set(contextGeneration, activePrompt) } + indexAssistantStepTiming(assistantSteps, event) if (event.type === 'request/header') { activeRequestConfig = event.data.header.config activePrompt = { @@ -406,30 +386,10 @@ export function projectConversationHistory( tools: event.data.header.tools ?? [], } promptsByContext.set(contextGeneration, activePrompt) - } else if (event.type === 'step/start') { - assistantSteps.set( - assistantStepKey(event.data.turn, event.data.step), - { stepStartTime: event.time, firstTokenTime: null }, - ) - } else if (event.type === 'assistant/chunk' && isTokenDelta(event.data.chunk)) { - const key = assistantStepKey(event.data.turn, event.data.step) - const current = assistantSteps.get(key) ?? { - stepStartTime: null, - firstTokenTime: null, - } - if (current.firstTokenTime === null) { - assistantSteps.set(key, { ...current, firstTokenTime: event.time }) - } } else if (event.type === 'assistant/message') { assistantTimings.set( event.seq, - { - ...(assistantSteps.get(assistantStepKey(event.data.turn, event.data.step)) ?? { - stepStartTime: null, - firstTokenTime: null, - }), - completedTime: event.time, - }, + settledAssistantTiming(assistantSteps, event.data.turn, event.data.step, event.time), ) if (activeRequestConfig !== undefined) { assistantRequestConfigs.set(event.seq, activeRequestConfig) diff --git a/packages/client/runtime/src/client/sessions/assistant-timing.ts b/packages/client/runtime/src/client/sessions/assistant-timing.ts new file mode 100644 index 0000000000..021c679d04 --- /dev/null +++ b/packages/client/runtime/src/client/sessions/assistant-timing.ts @@ -0,0 +1,84 @@ +// Shared assistant step-timing fold: both transcript projections (the live +// window adapter and the trajectory history fold) derive AssistantTiming from +// the same step/start -> first token delta -> assistant/message sequence, so +// the derivation lives once here instead of drifting per projection. + +import type { SessionEvent } from '@deepseek-ai/dsh-session/types' +import type { AssistantTiming } from './conversation.ts' + +/** Pre-finalize timing boundaries for one assistant step (start + first token). */ +export interface AssistantStepMetadata { + stepStartTime: number | null + firstTokenTime: number | null +} + +/** + * Composite map key for one assistant step. + * @param turn - turn number from the event payload. + * @param step - step number from the event payload. + * @returns collision-free `turn`/`step` key (NUL separator). + */ +export function assistantStepKey(turn: number, step: number): string { + return `${turn}\u0000${step}` +} + +/** + * Whether a chunk carries visible model output (first-token boundary). Empty + * deltas (heartbeats, empty tool-call frames) do not count as a first token. + * @param chunk - the assistant/chunk payload. + * @returns true when the chunk contains a non-empty text/reasoning/tool delta. + */ +export function isTokenDelta(chunk: SessionEvent<'assistant/chunk'>['data']['chunk']): boolean { + switch (chunk.type) { + case 'text-delta': + case 'reasoning-delta': + return chunk.text !== '' + case 'tool-call-delta': + return chunk.argumentsDelta !== '' || chunk.name !== undefined + default: + return false + } +} + +/** + * Fold one event into the per-step timing index: step/start opens the entry, + * the first non-empty token delta stamps first-token time once. Other event + * types are no-ops. + * @param steps - the mutable per-step index, keyed by {@link assistantStepKey}. + * @param event - the raw window event. + */ +export function indexAssistantStepTiming(steps: Map, event: SessionEvent): void { + if (event.type === 'step/start') { + steps.set( + assistantStepKey(event.data.turn, event.data.step), + { stepStartTime: event.time, firstTokenTime: null }, + ) + } else if (event.type === 'assistant/chunk' && isTokenDelta(event.data.chunk)) { + const key = assistantStepKey(event.data.turn, event.data.step) + const current = steps.get(key) ?? { stepStartTime: null, firstTokenTime: null } + if (current.firstTokenTime === null) { + steps.set(key, { ...current, firstTokenTime: event.time }) + } + } +} + +/** + * Settle one finalized assistant message's timing from its step entry; a step + * whose start or first token fell outside the window yields null boundaries. + * @param steps - the per-step index built by {@link indexAssistantStepTiming}. + * @param turn - the assistant/message turn number. + * @param step - the assistant/message step number. + * @param completedTime - the assistant/message event timestamp (epoch ms). + * @returns the node-ready timing record. + */ +export function settledAssistantTiming( + steps: ReadonlyMap, + turn: number, + step: number, + completedTime: number, +): AssistantTiming { + return { + ...(steps.get(assistantStepKey(turn, step)) ?? { stepStartTime: null, firstTokenTime: null }), + completedTime, + } +} diff --git a/packages/client/runtime/src/client/sessions/transcript-adapter.ts b/packages/client/runtime/src/client/sessions/transcript-adapter.ts index f9dc2d7871..306571b2bf 100644 --- a/packages/client/runtime/src/client/sessions/transcript-adapter.ts +++ b/packages/client/runtime/src/client/sessions/transcript-adapter.ts @@ -24,6 +24,8 @@ import type { CommandNode, CompactionSummaryNode, ConversationNode } from './con import { toAssistantBlocks } from './conversation.ts' import { contextForm, contextProvenance } from './context-provenance.ts' import { SteeringHistory } from './steering-history.ts' +import type { AssistantStepMetadata } from './assistant-timing.ts' +import { indexAssistantStepTiming, settledAssistantTiming } from './assistant-timing.ts' /** * The compaction seam's checkpoint plugin, pinned to the seam's own declaration @@ -52,6 +54,7 @@ function materializeNode( callIndex: ReadonlyMap, resultView: ToolResultView | null, steering: boolean, + stepTimings: ReadonlyMap, ): ConversationNode { switch (event.type) { case 'user/message': @@ -82,6 +85,7 @@ function materializeNode( kind: 'assistant', seq: event.seq, time: event.time, turn: event.data.turn, step: event.data.step, blocks: toAssistantBlocks(event.data.message.content), usage: event.data.usage, + timing: settledAssistantTiming(stepTimings, event.data.turn, event.data.step, event.time), } case 'tool/result': { const result = event.data.message.content[0] @@ -182,6 +186,8 @@ export class TranscriptAdapter { /** Transcript nodes in log order; copy-on-write so a published array never mutates. */ private projected: ConversationNode[] = [] private callIdx = new Map() + /** Per-step timing boundaries (step/start + first token delta), consumed when the step's assistant/message materializes. */ + private stepTimings = new Map() /** Wire result views keyed by the tool/result event's seq (views ride the envelope, not the event). */ private resultViews = new Map() /** Durable inbox replay used to distinguish next-step human input from queued prompts. */ @@ -216,6 +222,7 @@ export class TranscriptAdapter { this.commandIdx = new Map() this.steeringHistory.reset() const steeringSeqs = new Set() + this.stepTimings = new Map() for (let i = 0; i < events.length; i++) { const event = events[i] /* v8 ignore next -- dense-array guard: i stays within events.length, so the undefined arm needs a sparse array no caller builds. */ @@ -224,6 +231,7 @@ export class TranscriptAdapter { this.indexCall(event, views?.[i]) this.indexCommand(event) if (this.steeringHistory.apply(event)) steeringSeqs.add(event.seq) + indexAssistantStepTiming(this.stepTimings, event) } // Indexes first, then project: a tool/result materializes against the // complete call index, and a checkpoint against the complete event index. @@ -247,6 +255,7 @@ export class TranscriptAdapter { this.eventIndex.set(event.seq, event) this.indexCall(event, view) const steering = this.steeringHistory.apply(event) + indexAssistantStepTiming(this.stepTimings, event) if (this.indexCommand(event)) this.rev++ if (!isTranscriptEvent(event)) return this.projected = [...this.projected, this.materialize(event, steering)] @@ -285,7 +294,13 @@ export class TranscriptAdapter { private materialize(event: SessionEvent, steering: boolean): ConversationNode { return isCompactCheckpoint(event) ? materializeCompaction(event, this.eventIndex) - : materializeNode(event, this.callIdx, this.resultViews.get(event.seq) ?? null, steering) + : materializeNode( + event, + this.callIdx, + this.resultViews.get(event.seq) ?? null, + steering, + this.stepTimings, + ) } /** diff --git a/packages/client/runtime/tests/transcript-adapter.spec.ts b/packages/client/runtime/tests/transcript-adapter.spec.ts index a8b29d83b5..031acf1780 100644 --- a/packages/client/runtime/tests/transcript-adapter.spec.ts +++ b/packages/client/runtime/tests/transcript-adapter.spec.ts @@ -476,4 +476,48 @@ describe('TranscriptAdapter', () => { expect(nodes[1]).toMatchObject({ name: 'compact', outcome: { kind: 'success', text: '已压缩' } }) }) }) + + describe('assistant timing', () => { + const base = 1_700_000_000_000 + + it('derives step timing across a window rebuild (start + first token + completion)', () => { + const adapter = new TranscriptAdapter() + adapter.reset([ + ev.turnStart(0, 0), + ev.user(1, '问'), + ev.stepStart(2, 0), + ev.chunkStart(3, 0), + ev.chunkText(4, 0, '答'), + ev.chunkText(5, 0, '案'), + ev.assistant(6, 0, '答案'), + ev.turnEnd(7, 0), + ]) + const assistant = adapter.nodes().find(n => n.kind === 'assistant') + expect(assistant).toMatchObject({ + timing: { stepStartTime: base + 2, firstTokenTime: base + 4, completedTime: base + 6 }, + }) + }) + + it('derives the same timing on the live append path, first token winning once', () => { + const adapter = new TranscriptAdapter() + adapter.reset([ev.user(0, '问')]) + adapter.append(ev.stepStart(1, 0)) + adapter.append(ev.chunkText(2, 0, '首')) + adapter.append(ev.chunkText(3, 0, '次')) + adapter.append(ev.assistant(4, 0, '首次')) + const assistant = adapter.nodes().find(n => n.kind === 'assistant') + expect(assistant).toMatchObject({ + timing: { stepStartTime: base + 1, firstTokenTime: base + 2, completedTime: base + 4 }, + }) + }) + + it('soft-falls to null boundaries when the step opening fell outside the window', () => { + const adapter = new TranscriptAdapter() + adapter.reset([ev.assistant(100, 0, '被切窗的答案')]) + const assistant = adapter.nodes().find(n => n.kind === 'assistant') + expect(assistant).toMatchObject({ + timing: { stepStartTime: null, firstTokenTime: null, completedTime: base + 100 }, + }) + }) + }) }) diff --git a/packages/client/ui-conversation/README.i18n.yaml b/packages/client/ui-conversation/README.i18n.yaml index ff7ec73725..0df2b4b4df 100644 --- a/packages/client/ui-conversation/README.i18n.yaml +++ b/packages/client/ui-conversation/README.i18n.yaml @@ -2,5 +2,5 @@ # side as of the last confirmed-consistent state. Both languages carry equal authority; # after editing either side, bring the other along and re-record with: # pnpm run verify-translation-pairing --write packages/client/ui-conversation/README.md -README.md: 2e71fcb899b38b0d6d8a37a83444aa4a930c365d -README.zh.md: c683c8295980e773c934bd323a3e01a960bccc75 +README.md: 7bd0d551fc41967326dd9860f5c31a99ea3c254a +README.zh.md: d339f6423d9a9f77c02d86ad0b8e57bd0baba52b diff --git a/packages/client/ui-conversation/README.md b/packages/client/ui-conversation/README.md index 2e71fcb899..7bd0d551fc 100644 --- a/packages/client/ui-conversation/README.md +++ b/packages/client/ui-conversation/README.md @@ -46,7 +46,7 @@ Per-session UI state for selection and the active view lives in the declared cha The composer bar declares session-scoped single seats for `'conversation.input.plan'` (right of the local access-mode control) and `'conversation.input.model'` (immediately before the pending indicator and send/stop button), plus list slots for overlay, dock, left, and right input extensions. Feature packages own each control and its state; ui-conversation supplies placement, the `locked` owner prop, and the standard slot shares. The leading plus button is a Command launcher, not an attachment surface: it asks the session's `SlashController` to open only the `/` trigger's `command` source over the current textarea selection, while ui-slash's existing `MenuView` remains the sole floating menu and pick path. No file row, file input, upload protocol, or second menu component is introduced. While the `plan` projection's effective target is plan mode, InputBar swaps its textarea placeholder to the plan-task wording, localized through the `conversation` locale namespace this package registers (the `placeholder.plan` / `hint.plan` keys) and shared verbatim with the claimed `/plan` command hint (a host-folded value read through the standard-kit `useProjection`; owner-supplied placeholders win). A pending composer takeover remains mounted when another conversation view is active so the blocked agent can still receive its answer; without a pending interaction, the active-session composer belongs to Chat. The composer-bar slot itself is `session-maybe`: with no current session the same bar renders inert (machine faces absent, `disabled` owner prop) instead of swapping in a parallel disabled tree, so the textarea DOM survives the workspace pick; the strict-session control seats simply stay empty until a session exists. -The chat stats line takes its token accounting from two generic token-meter projections read through the standard-kit `useProjection`: `tokenUsage` for full-log billing (billed input is uncached input plus cache reads and writes; cache hit divides cache reads by that total) and `contextPressure` for context occupancy. Visible nodes supply only the turn and step counts plus the LLM and tool wall times, which are window-scoped facts about what is on screen rather than accounting; durable token and context groups remain visible when compaction leaves no assistant node in the loaded window. A deployment without token-meter drops the token groups, and occupancy stays hidden until both provider pressure and route capacity are known. Occupancy is deliberately an approximation — its numerator and capacity are independent last-wins projection fields, not one atomic request observation ([rationale](../../llm/token-meter/README.md)). The inline stats row remains the sole context UI; the model selector has no circle or accessory. +The chat stats line takes its token accounting from the generic token-meter `tokenUsage` projection read through the standard-kit `useProjection`: billed input is uncached input plus cache reads and writes; cache hit divides cache reads by that total. Visible nodes supply only the turn and step counts plus the LLM and tool wall times, which are window-scoped facts about what is on screen rather than accounting; durable token and context groups remain visible when compaction leaves no assistant node in the loaded window. The same window fold averages each recorded step's TTFT and divides sampled output tokens by their summed decode spans into a latency/throughput group localized through the `conversation` locale namespace (`TTFT avg … · … tok/s` in English); a step missing a timing boundary or a usage sample drops out of those figures instead of skewing them. The turn-count, step-count, duration, cache, and token labels use the same namespace. Each settled turn additionally appends hover-revealed `TTFT {s}s · {tps} tok/s` labels to its assistant footer after the `Ran for` duration — the turn's first-step TTFT and its turn-aggregate decode throughput — gated on the turn's timing being in the loaded window (a contiguous log suffix, so an in-window turn carries every one of its steps) and omitting whichever figure is unrecorded. A deployment without token-meter drops the token groups; when the line overflows, it elides with an ellipsis and a delayed hover tooltip carries the full text only while actually clipped. Context occupancy moved off the row onto the composer's trailing ContextMeter: a 14px occupancy ring after the model seat, fed by `contextPressure` and rendered only once both a numerator and a route capacity are known, that click-opens a panel pairing the `percent used` header and `~used / capacity` figures with a color-segmented bar and `~`-prefixed heuristic composition rows (system prompt, tools, messages) from the `contextBreakdown` projection. The ring and header read `projectedTokens` — the provider sample carried forward over the surface's movement since — so a compaction registers immediately instead of after a further turn; the composition rows stay wholly heuristic and therefore still do not sum to the header ([rationale](../../llm/token-meter/README.md)). Occupancy is deliberately an approximation: numerator and capacity are independent last-wins projection fields, not one atomic request observation. `src/client/` is organized by domain. `contract/` is the sole inter-domain shared face (`slots.ts` slot declarations and composed props, `views.ts` shared primitives, `tool-call-model.ts`); the `skeleton/`, `chat/`, and `toolviews/` directories import contract files and never each other. `apply.ts` is the only assembly point allowed to import all three domains. The `/client` export surface is the contract only — `apply`/`inject`, the two service classes, and the `contract/` type families; implementation components and the store factory stay internal and reach the page through apply's slot registrations. @@ -61,7 +61,7 @@ None; this package neither assembles nor sends a provider request. ## Known Limitations and Deferred Work - **Compaction markers show no scale** — the row does not yet report how many messages or which range the checkpoint replaced. -- **Stats-line durations cover the in-window flow only** — LLM and tool wall times fold the snapshot's assistant `timing` and tool call/result pairs, so nodes outside the loaded event window (older history) are not counted. +- **Stats-line durations and speeds cover the in-window flow only** — LLM and tool wall times plus the TTFT and throughput averages fold the snapshot's assistant `timing` and tool call/result pairs, so nodes outside the loaded event window (older history) are not counted. - **The details panel has no entry point** — `ChatViewInjected.openDetails` is implemented but uncalled, so the raw selected-call display is unreachable in the assembled application. There is no Input/Output/Metadata switch, Prev/Next stepping, or trajectory deep link. - **Assistant per-message paging is a reserved slot** — drawn in the design, not implemented. The finalized content IconActions row (copy / clock / branch) ships under the last content-text assistant of each turn only; mid-turn narration and Think-only nodes stay chrome-free. Branch stays disabled unless that message is also the last transcript node of a completed turn; when enabled, it forks through that turn, increments the inherited title on the client, and opens the child. A fork or rename failure leaves the source selected ([decision](../../../.agents/notes/implemented/bug-fix/2026-08-02-message-fork-actions-require-completed-turn-tail.md)). - **Sent user messages cannot be edited** — user bubbles retain clock, copy, and branch; branch stays disabled unless a completed turn's transcript ends at that user message. Editing returns with the capability behind it: a client mutation over a settled user message, plus the host behavior for the turn that already consumed it ([decision](../../../.agents/notes/implemented/simplification/2026-07-31-drop-user-message-edit-stub.md)). diff --git a/packages/client/ui-conversation/README.zh.md b/packages/client/ui-conversation/README.zh.md index c683c82959..d339f6423d 100644 --- a/packages/client/ui-conversation/README.zh.md +++ b/packages/client/ui-conversation/README.zh.md @@ -46,9 +46,9 @@ Host 带 placement 的 `session/queue` 快照也会携带待处理 steering。Qu 输入栏为 `'conversation.input.plan'`（位于本地 access 模式控件右侧）和 `'conversation.input.model'`（渲染在 pending 指示器与发送／停止按钮之前）声明会话作用域的单实例 seat，并为 overlay、dock、left 和 right 输入扩展声明列表 slot。各功能包拥有相应控件及其状态；ui-conversation 提供放置位置、`locked` owner prop 和标准 slot share。前置加号按钮是 Command launcher，而非附件入口：它要求当前会话的 `SlashController` 基于 textarea 当前 selection，只打开 `/` trigger 的 `command` source，同时 ui-slash 既有的 `MenuView` 仍是唯一的浮层菜单与 pick 路径。不引入 File 行、file input、上传协议或第二套菜单组件。当 `plan` 投影的有效目标为 plan mode 时，InputBar 将文本框 placeholder 切换为 plan 任务措辞，经本包注册的 `conversation` locale 命名空间（`placeholder.plan` / `hint.plan` 键）本地化，并与已认领 `/plan` 命令的提示逐字共用同一份文案（经标准套件 `useProjection` 读取的 host 折叠值；owner 提供的 placeholder 优先）。另一个会话视图活跃时，待处理的 composer 接管仍保持挂载，使被阻塞的 agent（智能体）仍能收到回答；没有待处理交互时，活跃会话的 composer 归 Chat 所有。composer bar slot 本身为 `session-maybe`：没有当前会话时，同一个 bar 以不可交互状态渲染（machine face 均缺席、`disabled` owner prop），而不是换入一棵平行的 disabled 树，因此选择 workspace 时 textarea DOM 不会被销毁；严格会话作用域的控件 seat 在会话存在之前保持为空。 -聊天统计行的 token 账目来自经标准套件 `useProjection` 读取的两个通用 token-meter 投影：`tokenUsage` 提供完整日志计费用量（计费输入为未缓存输入、缓存读取与缓存写入之和；缓存命中率以缓存读取除以该总量），`contextPressure` 提供上下文占用率。可见节点只提供轮次与步骤计数，以及 LLM（大语言模型）和工具的墙钟时间：这些是关于「屏幕上有什么」的窗口作用域事实，而非账目；压缩（compaction）使已加载窗口不再包含 assistant 节点时，持久 token 与上下文分组仍保持可见。未组合 token-meter 的部署会整组省略 token 分组；只有提供方压力与路由容量都已知时才显示占用率。占用率是刻意为之的近似值：它的分子与容量是两个相互独立的「后写覆盖」投影字段，并非同一次请求的原子观测（[原理](../../llm/token-meter/README.md)）。行内统计行仍是唯一的上下文 UI；模型选择器不增加圆环或附属控件。 +聊天统计行的 token 账目来自经标准套件 `useProjection` 读取的通用 token-meter 投影 `tokenUsage`：计费输入为未缓存输入、缓存读取与缓存写入之和；缓存命中率以缓存读取除以该总量。可见节点只提供轮次与步骤计数，以及 LLM（大语言模型）和工具的墙钟时间：这些是关于「屏幕上有什么」的窗口作用域事实，而非账目；压缩（compaction）使已加载窗口不再包含 assistant 节点时，持久 token 与上下文分组仍保持可见。同一次窗口折算还会把每个有完整记录的步骤的 TTFT（首 token 延迟）取平均，并用采样到的输出 token 数除以其解码时长之和，得到经 `conversation` locale 命名空间本地化的延迟／吞吐分组（中文为 `首 token 平均 … · … tok/s`）；缺少某个 timing 边界或 usage 采样的步骤会直接退出这些数字，而不是让它们失真。轮次计数、步骤计数、耗时、缓存与 token 各项的标签也使用同一命名空间。每个已结算轮次还会在其 assistant footer 的 `用时` 之后追加 hover 才显示的 `首 token {s}秒 · {tps} tok/s` 标签——即该轮次首个步骤的 TTFT 与轮次聚合的解码吞吐——仅当该轮次的 timing 位于已加载窗口内才显示（窗口是日志的连续后缀，因此窗口内的轮次必然带着它的全部步骤），未记录的数字会各自省略。未组合 token-meter 的部署会整组省略 token 分组；统计行过长时以省略号截断，仅在内容真的被裁切时由延迟 hover tooltip 承载完整文本。上下文占用率从统计行移到了 composer 尾部的 ContextMeter：模型座位之后的一枚 14px 占用圆环，由 `contextPressure` 供数，仅当分子与路由容量都已知时才渲染；点击弹出的面板把「已用百分比」标题与 `~已用 / 容量` 数字，与来自 `contextBreakdown` 投影、带 `~` 前缀的启发式组成明细行（系统提示词、工具、对话消息）及分色分段进度条并列。圆环与标题读取 `projectedTokens`——把提供方样本沿此后表层的增减推进到当下——因此压缩会立刻反映出来，而不必再等一整轮；组成明细行仍是纯启发式，因此加起来依然不等于标题数字（[原理](../../llm/token-meter/README.md)）。占用率是刻意为之的近似值：分子与容量是两个相互独立的「后写覆盖」投影字段，并非同一次请求的原子观测。 -`src/client/` 按未来的包拆分组织：`contract/` 是唯一的跨领域共享表层（`slots.ts` slot 声明 + 组合后的 slot props，包括工具行契约、`views.ts` 共享原语、`tool-call-model.ts`）；`skeleton/`、`chat/` 和 `toolviews/`（示例注册方）领域目录只导入 contract 文件，彼此绝不导入；`apply.ts` 是唯一允许导入全部三个领域的组装点。`/client` 导出表层只包含契约：`apply`／`inject`、两个服务类和 `contract/` 类型家族；实现组件（骨架、聊天行）与 store factory 保持内部状态，只能通过 apply 的 slot 注册到达页面（测试通过 `./src/*` 子路径获取它们）。 +`src/client/` 按领域组织。`contract/` 是唯一的跨领域共享表层（`slots.ts` slot 声明与组合后的 props、`views.ts` 共享原语、`tool-call-model.ts`）；`skeleton/`、`chat/` 和 `toolviews/` 目录只导入 contract 文件，彼此之间从不互相导入。`apply.ts` 是唯一允许导入全部三个领域的组装点。`/client` 导出表层只包含契约：`apply`／`inject`、两个服务类和 `contract/` 类型家族；实现组件与 store factory 保持内部，经 apply 的 slot 注册抵达页面。 ## 模型体验 @@ -61,8 +61,8 @@ Host 带 placement 的 `session/queue` 快照也会携带待处理 steering。Qu ## 已知限制与暂缓事项 - **压缩标记不显示规模**：该行尚不报告检查点替换了多少条消息或哪段范围。 -- **统计行的耗时只覆盖窗口内消息流**：LLM 与工具墙钟时间由快照的 assistant `timing` 与工具 call/result 配对折算，落在已加载事件窗口之外的节点（更早的历史）不计入。 -- **详情面板是最小形态，且当前没有入口**：以原始形式显示已选择调用的参数／结果；Input/Output/Metadata 切换、Prev/Next 步进与 See-in-trajectory 深链接暂缓实现。工具行已不再是详情面板的点击目标，且没有任何手势接替它，因此 `ChatViewInjected.openDetails` 虽已实现却无人调用，该面板（含其终端卡片）在组装后的应用中不可达；其渲染仍由直接以选中态挂载它来覆盖。 +- **统计行的耗时与速率只覆盖窗口内消息流**：LLM 与工具墙钟时间以及 TTFT 与吞吐平均值由快照的 assistant `timing` 与工具 call/result 配对折算，落在已加载事件窗口之外的节点（更早的历史）不计入。 +- **详情面板没有入口**：`ChatViewInjected.openDetails` 虽已实现却无人调用，因此以原始形式显示已选择调用的那部分在组装后的应用中不可达。没有 Input/Output/Metadata 切换、Prev/Next 步进，也没有 trajectory 深链接。 - **assistant 逐消息分页是预留 slot**：设计中已有图稿，尚未实现。已定稿的内容 IconActions 行（复制／时钟／分支）只挂在每个轮次中最后一条带 text 内容的 assistant 下；轮次中间的叙述与纯 Think 节点不带 chrome。除非该消息同时也是已完成轮次的最后一个 transcript 节点，否则分支保持禁用；启用后，它会 fork 到该轮次末尾，在 client 端递增继承标题并打开子会话。fork 或改名失败时源会话保持选中（[决策](../../../.agents/notes/implemented/bug-fix/2026-08-02-message-fork-actions-require-completed-turn-tail.md)）。 - **已发送的 user 消息无法编辑**：user 气泡保留时钟、复制和分支；除非已完成轮次的 transcript 结束于该 user 消息，否则分支保持禁用。编辑功能要与其背后的能力一起回归：既需要针对已定稿 user 消息的 client 变更，也需要 host 侧对已经消费过它的轮次给出行为（[决策](../../../.agents/notes/implemented/simplification/2026-07-31-drop-user-message-edit-stub.md)）。 - **others 工具行的闪光图标是手绘近似版本**：无法在本地导出设计字形的矢量几何；等到存在精确导出后再将其提升到 ui-primitives。 diff --git a/packages/client/ui-conversation/src/client/apply.ts b/packages/client/ui-conversation/src/client/apply.ts index e6a38e491f..b796ce84fc 100644 --- a/packages/client/ui-conversation/src/client/apply.ts +++ b/packages/client/ui-conversation/src/client/apply.ts @@ -330,7 +330,7 @@ export function apply(ctx: Context): void { }, ChatView) // Session stats stick with the composer (composer.dock = stats-line family). - slots.register({ name: 'conversation.composer.dock', id: 'stats', order: 0 }, StatsLine) + slots.register({ name: 'conversation.composer.dock', id: 'stats', order: 0, locale: NS }, StatsLine) // Class-plugin mount (packages/AGENTS.md service form): the service // registers itself as `conversation` and lives on its own child fiber. diff --git a/packages/client/ui-conversation/src/client/chat/AssistantMarkdown.tsx b/packages/client/ui-conversation/src/client/chat/AssistantMarkdown.tsx index de882e8bb2..5b3b9fa821 100644 --- a/packages/client/ui-conversation/src/client/chat/AssistantMarkdown.tsx +++ b/packages/client/ui-conversation/src/client/chat/AssistantMarkdown.tsx @@ -30,6 +30,10 @@ export interface AssistantMarkdownProps { /** Turn wall time in ms for the IconActions run-time label; omitted when the * turn's triggering input is outside the loaded window. */ runMs?: number | undefined + /** Turn first-step TTFT in ms for the IconActions label; omitted when unrecorded. */ + ttftMs?: number | undefined + /** Turn decode throughput for the IconActions label; omitted when unrecorded. */ + tokensPerSecond?: number | undefined /** Event sequence used as the fork boundary; omitted while streaming. */ seq?: number | undefined /** Fork the session through this finalized message's completed turn when eligible. */ @@ -82,7 +86,7 @@ function ThinkRow({ text, running, t }: { text: string; running: boolean; t: Ass } export const AssistantMarkdown = memo(function AssistantMarkdown({ - blocks, streaming, interrupted, time, runMs, seq, onFork, forkUnavailable, t, + blocks, streaming, interrupted, time, runMs, ttftMs, tokensPerSecond, seq, onFork, forkUnavailable, t, }: AssistantMarkdownProps) { // Stable per locale revision (t identity changes on switch): a fresh object // per render would rebuild MarkdownText's component table every chunk. @@ -125,6 +129,8 @@ export const AssistantMarkdown = memo(function AssistantMarkdown({ text={copyText(blocks)} time={time} runMs={runMs} + ttftMs={ttftMs} + tokensPerSecond={tokensPerSecond} clock="end" onBranch={onFork === undefined || seq === undefined ? undefined : () => { onFork(seq) }} branchUnavailable={forkUnavailable} diff --git a/packages/client/ui-conversation/src/client/chat/ChatView.tsx b/packages/client/ui-conversation/src/client/chat/ChatView.tsx index 3320eb0f69..c852161240 100644 --- a/packages/client/ui-conversation/src/client/chat/ChatView.tsx +++ b/packages/client/ui-conversation/src/client/chat/ChatView.tsx @@ -36,6 +36,7 @@ import { GenericCommandCard } from './GenericCommandCard.tsx' import { GenericToolCard } from './GenericToolCard.tsx' import { MessageItem, PendingSteeringBubble } from './MessageItem.tsx' import { formatRunDuration } from './message-chrome.ts' +import { deriveTurnMetrics } from './turn-metrics.ts' import css from './ChatView.module.css' const FOLLOW_THRESHOLD = 24 @@ -362,6 +363,7 @@ export function ChatView({ const actionSeqs = useMemo(() => assistantActionsSeqs(nodes), [nodes]) const branchSeqs = useMemo(() => messageBranchSeqs(nodes, turnEnds), [nodes, turnEnds]) const runningTurnStart = useMemo(() => runningTurnStartTime(turnTimings), [turnTimings]) + const turnMetrics = useMemo(() => deriveTurnMetrics(nodes), [nodes]) const listRef = useRef(null) const columnRef = useRef(null) @@ -599,6 +601,9 @@ export function ChatView({ const node: ConversationNode = item.node if (node.kind === 'assistant') { const timing = actionSeqs.has(node.seq) ? turnTimings.get(node.turn) : undefined + // Metrics gate on the settled in-window timing: turn/start loaded means + // every step of the turn is loaded, so first-step TTFT is genuine. + const metrics = timing?.endTime === undefined ? undefined : turnMetrics.get(node.turn) return ( {formatMessageClock(time, t, day)} {runMs !== undefined && ( <> + {' '} · + {' '} {t('message.ranFor', { duration: formatRunDuration(runMs, t) })} )} + {ttftMs !== undefined && ( + <> + {' '} + · + {' '} + {t('message.ttft', { seconds: formatLatencySeconds(ttftMs) })} + + )} + {tokensPerSecond !== undefined && ( + <> + {' '} + · + {' '} + {t('message.tokensPerSecond', { tps: formatTokensPerSecond(tokensPerSecond) })} + + )} ) return ( diff --git a/packages/client/ui-conversation/src/client/chat/StatsLine.tsx b/packages/client/ui-conversation/src/client/chat/StatsLine.tsx index ceffffb59a..8e672740cf 100644 --- a/packages/client/ui-conversation/src/client/chat/StatsLine.tsx +++ b/packages/client/ui-conversation/src/client/chat/StatsLine.tsx @@ -2,10 +2,14 @@ // Mounted on 'conversation.composer.dock' so it sticks with the composer in the // active conversation scrollport (see ConversationRoot data-conversation-scroll). -import { Fragment, memo, useMemo } from 'react' +import { Fragment, memo, useLayoutEffect, useMemo, useRef, useState } from 'react' +import { Tooltip } from '@deepseek-ai/dsh-client-ui-primitives' import type { ConversationSnapshot, UseProjection } from '@deepseek-ai/dsh-client-runtime/client' import type { SnapshotSelectorHook } from '@deepseek-ai/dsh-client-ui-slots' import type { ContextPressureProjection, TokenUsageProjection } from '@deepseek-ai/dsh-token-meter/client' +import type { ComposerBarProps } from '../contract/slots.ts' +import { formatTokensPerSecond } from './message-chrome.ts' +import { assistantStepReading } from './turn-metrics.ts' import css from './StatsLine.module.css' interface WindowStats { @@ -15,6 +19,14 @@ interface WindowStats { llmMs: number /** Summed tool wall time (tool/call → tool/result); 0 when no pair is in-window. */ toolMs: number + /** Summed first-token latency over `ttftSteps`; 0 when no step records it. */ + ttftMs: number + /** Steps carrying a recorded TTFT. */ + ttftSteps: number + /** Summed decode wall time over steps that also report output tokens. */ + decodeMs: number + /** Summed output tokens over the same decode-timed steps. */ + decodeTokens: number } /** @@ -32,6 +44,10 @@ export function deriveStats(nodes: ConversationSnapshot['nodes']): WindowStats { let steps = 0 let llmMs = 0 let toolMs = 0 + let ttftMs = 0 + let ttftSteps = 0 + let decodeMs = 0 + let decodeTokens = 0 for (const node of nodes) { if (node.kind === 'tool-result') { if (node.callTime !== null) toolMs += Math.max(0, node.time - node.callTime) @@ -43,8 +59,17 @@ export function deriveStats(nodes: ConversationSnapshot['nodes']): WindowStats { if (node.timing !== undefined && node.timing.stepStartTime !== null) { llmMs += Math.max(0, node.timing.completedTime - node.timing.stepStartTime) } + const reading = assistantStepReading(node) + if (reading.ttftMs !== null) { + ttftMs += reading.ttftMs + ttftSteps += 1 + } + if (reading.decodeMs !== null && reading.outputTokens !== null) { + decodeMs += reading.decodeMs + decodeTokens += reading.outputTokens + } } - return { turns: turns.size, steps, llmMs, toolMs } + return { turns: turns.size, steps, llmMs, toolMs, ttftMs, ttftSteps, decodeMs, decodeTokens } } /** @@ -84,30 +109,41 @@ export function cacheHitPercent(usage: TokenUsageProjection): number | null { : Math.round(usage.cacheReadTokens / denominator * 100) } -/** Sum the three disjoint prompt-side billing buckets. */ -function billedInputTokens(usage: TokenUsageProjection): number { +/** + * Sum the three disjoint prompt-side billing buckets. + * @param usage - the session's token-usage projection value. + * @returns billed input tokens. + */ +export function billedInputTokens(usage: TokenUsageProjection): number { return usage.uncachedInputTokens + usage.cacheReadTokens + usage.cacheWriteTokens } interface ContextOccupancy { percent: number + usedTokens: number contextWindow: number } /** * Approximate context occupancy, using the TUI's integer rounding and upper - * clamp. The numerator and capacity are independent last-wins projection - * fields, so this is a reference figure rather than an exact measurement of one - * request (see the token-meter README). + * clamp. The numerator is `projectedTokens` — the provider sample carried + * forward over the surface's movement since — so compaction shows immediately + * instead of waiting for the next request to report usage; it falls back to the + * bare sample only for a log whose projection predates that field. Numerator + * and capacity remain independent last-wins projection fields, so this is a + * reference figure rather than an exact measurement of one request (see the + * token-meter README). * @param pressure - the session's context-pressure projection value. - * @returns occupancy and its denominator, or null until both values are known. + * @returns occupancy with its numerator and denominator, or null until both values are known. */ export function contextOccupancy( pressure: ContextPressureProjection | undefined, ): ContextOccupancy | null { - if (pressure?.pressureTokens === undefined || pressure.contextWindow === undefined) return null + const usedTokens = pressure?.projectedTokens ?? pressure?.pressureTokens + if (usedTokens === undefined || pressure?.contextWindow === undefined) return null return { - percent: Math.min(100, Math.round(pressure.pressureTokens / pressure.contextWindow * 100)), + percent: Math.min(100, Math.round(usedTokens / pressure.contextWindow * 100)), + usedTokens, contextWindow: pressure.contextWindow, } } @@ -116,46 +152,73 @@ export function contextOccupancy( export interface StatsLineProps { useSession: SnapshotSelectorHook useProjection: UseProjection + /** The owning dock's locale seat. */ + t: ComposerBarProps['t'] } -export const StatsLine = memo(function StatsLine({ useSession, useProjection }: StatsLineProps) { +export const StatsLine = memo(function StatsLine({ useSession, useProjection, t }: StatsLineProps) { const nodes = useSession(s => s.nodes) const usage = useProjection('tokenUsage') - const pressure = useProjection('contextPressure') const stats = useMemo(() => deriveStats(nodes), [nodes]) // Pipe-separated groups (figma stats strip); a group with no data drops out whole. const groups: string[] = [] if (stats.steps > 0) { - groups.push(`${stats.turns} turns · ${stats.steps} steps`) + groups.push(t('stats.counts', { turns: stats.turns, steps: stats.steps })) const durations: string[] = [] - if (stats.llmMs > 0) durations.push(`LLM ${formatDuration(stats.llmMs)}`) - if (stats.toolMs > 0) durations.push(`Tool call ${formatDuration(stats.toolMs)}`) + if (stats.llmMs > 0) durations.push(t('stats.llm', { duration: formatDuration(stats.llmMs) })) + if (stats.toolMs > 0) durations.push(t('stats.toolCall', { duration: formatDuration(stats.toolMs) })) if (durations.length > 0) groups.push(durations.join(' · ')) + // Window-scoped like the wall times above: averages describe loaded steps. + const speeds: string[] = [] + if (stats.ttftSteps > 0) { + speeds.push(t('stats.ttftAverage', { duration: formatDuration(stats.ttftMs / stats.ttftSteps) })) + } + if (stats.decodeMs > 0) { + speeds.push(t('stats.tokensPerSecond', { + throughput: formatTokensPerSecond(stats.decodeTokens / (stats.decodeMs / 1_000)), + })) + } + if (speeds.length > 0) groups.push(speeds.join(' · ')) } - const context = contextOccupancy(pressure) - if (context !== null) { - groups.push(`Context ${context.percent}% of ${formatTokens(context.contextWindow)}`) - } + // Context occupancy deliberately lives on the composer's ContextMeter ring, + // not here — one home per fact. // Billing rides the durable projection, so these survive paging and // compaction. Suppress the empty projection on a brand-new session. if (usage !== undefined && (stats.steps > 0 || billedInputTokens(usage) > 0 || usage.outputTokens > 0)) { const cacheHit = cacheHitPercent(usage) - if (cacheHit !== null) groups.push(`Cache hit ${cacheHit}%`) - groups.push( - `Input ${formatTokens(billedInputTokens(usage))} tok` - + ` · Output ${formatTokens(usage.outputTokens)} tok`, - ) + if (cacheHit !== null) groups.push(t('stats.cacheHit', { percent: cacheHit })) + groups.push(t('stats.tokens', { + input: formatTokens(billedInputTokens(usage)), + output: formatTokens(usage.outputTokens), + })) } + const line = groups.join(' | ') + // The row elides with ellipsis when overlong; a delayed hover tooltip carries + // the full line, enabled only while content is actually clipped. + const rootRef = useRef(null) + const [truncated, setTruncated] = useState(false) + useLayoutEffect(() => { + const el = rootRef.current + if (el === null) return + const measure = () => { setTruncated(el.scrollWidth > el.clientWidth) } + measure() + if (typeof ResizeObserver === 'undefined') return + const observer = new ResizeObserver(measure) + observer.observe(el) + return () => { observer.disconnect() } + }, [line]) if (groups.length === 0) return null return ( -
- {groups.map((group, i) => ( - - {i > 0 && <>|{' '}} - {group} - - ))} -
+ +
+ {groups.map((group, i) => ( + + {i > 0 && <>|{' '}} + {group} + + ))} +
+ ) }) diff --git a/packages/client/ui-conversation/src/client/chat/message-chrome.ts b/packages/client/ui-conversation/src/client/chat/message-chrome.ts index 5837957011..a3658853c9 100644 --- a/packages/client/ui-conversation/src/client/chat/message-chrome.ts +++ b/packages/client/ui-conversation/src/client/chat/message-chrome.ts @@ -48,6 +48,27 @@ export function formatRunDuration(ms: number, t: RunDurationTranslate): string { : t('duration.seconds', { seconds }) } +/** + * Sub-turn latency figure: one decimal under ten seconds, whole seconds + * beyond. Unit-less so the locale template owns the second suffix. + * @param ms - Latency in milliseconds (negatives clamp to zero). + * @returns Display number in seconds without unit. + */ +export function formatLatencySeconds(ms: number): string { + const s = Math.max(0, ms) / 1000 + return s < 10 ? String(Math.round(s * 10) / 10) : String(Math.round(s)) +} + +/** + * Decode-throughput figure: whole tokens from ten up, one decimal below. + * @param tps - Tokens per second. + * @returns Display number without unit. + */ +export function formatTokensPerSecond(tps: number): string { + const clamped = Math.max(0, tps) + return clamped >= 10 ? String(Math.round(clamped)) : String(Math.round(clamped * 10) / 10) +} + /** * Compact local timestamp for message IconActions. Same calendar day → * `HH:mm`; earlier this year → the `clock.md` date template + clock; other diff --git a/packages/client/ui-conversation/src/client/chat/turn-metrics.ts b/packages/client/ui-conversation/src/client/chat/turn-metrics.ts new file mode 100644 index 0000000000..b7cc5ddb72 --- /dev/null +++ b/packages/client/ui-conversation/src/client/chat/turn-metrics.ts @@ -0,0 +1,97 @@ +// Latency/throughput folds shared by the settled turn footer and StatsLine. + +import type { ConversationSnapshot } from '@deepseek-ai/dsh-client-runtime/client' + +/** Latency and decode-throughput readings for one turn's footer. */ +export interface TurnMetrics { + /** First-step TTFT in ms; absent when that step carries no recorded timing. */ + ttftMs?: number + /** Decode throughput over steps carrying both timing and provider usage. */ + tokensPerSecond?: number +} + +/** One assistant step's derivable latency facts; null marks an unrecorded part. */ +export interface StepReading { + /** step/start → first token delta, in ms. */ + ttftMs: number | null + /** First token delta → final message, in ms. */ + decodeMs: number | null + /** Provider-reported completion tokens. */ + outputTokens: number | null +} + +interface UsageLike { + outputTokens?: number +} + +type AssistantNode = Extract + +function usageOutputTokens(usage: unknown): number | null { + if (typeof usage !== 'object' || usage === null) return null + const value = (usage as UsageLike).outputTokens + return typeof value === 'number' && Number.isFinite(value) && value >= 0 ? value : null +} + +/** + * Read one assistant node's TTFT, decode wall time, and output tokens. + * @param node - A settled assistant node. + * @returns Per-part readings with `null` for unrecorded values. + */ +export function assistantStepReading(node: AssistantNode): StepReading { + const timing = node.timing + const ttftMs = timing !== undefined && timing.stepStartTime !== null && timing.firstTokenTime !== null + ? Math.max(0, timing.firstTokenTime - timing.stepStartTime) + : null + const decodeMs = timing !== undefined && timing.firstTokenTime !== null + ? Math.max(0, timing.completedTime - timing.firstTokenTime) + : null + return { ttftMs, decodeMs, outputTokens: usageOutputTokens(node.usage) } +} + +interface TurnFold { + firstStep: number + firstStepTtftMs: number | null + decodeMs: number + outputTokens: number + sampled: boolean +} + +/** + * Fold assistant nodes into per-turn footer metrics. + * + * TTFT is the turn's lowest-step request-dispatch-to-first-token reading, so + * it is only meaningful when the turn's start is inside + * the loaded window (the caller gates on `turnTimings`, which shares that + * window). Throughput divides summed output tokens by summed decode wall time, + * counting only steps that carry both. + * @param nodes - Snapshot nodes of the loaded window. + * @returns Turn number → available metrics; turns with none are absent. + */ +export function deriveTurnMetrics(nodes: ConversationSnapshot['nodes']): Map { + const folds = new Map() + for (const node of nodes) { + if (node.kind !== 'assistant') continue + const reading = assistantStepReading(node) + let fold = folds.get(node.turn) + if (fold === undefined) { + fold = { firstStep: node.step, firstStepTtftMs: reading.ttftMs, decodeMs: 0, outputTokens: 0, sampled: false } + folds.set(node.turn, fold) + } else if (node.step < fold.firstStep) { + fold.firstStep = node.step + fold.firstStepTtftMs = reading.ttftMs + } + if (reading.decodeMs !== null && reading.outputTokens !== null) { + fold.decodeMs += reading.decodeMs + fold.outputTokens += reading.outputTokens + fold.sampled = true + } + } + const metrics = new Map() + for (const [turn, fold] of folds) { + const entry: TurnMetrics = {} + if (fold.firstStepTtftMs !== null) entry.ttftMs = fold.firstStepTtftMs + if (fold.sampled && fold.decodeMs > 0) entry.tokensPerSecond = fold.outputTokens / (fold.decodeMs / 1000) + if (entry.ttftMs !== undefined || entry.tokensPerSecond !== undefined) metrics.set(turn, entry) + } + return metrics +} diff --git a/packages/client/ui-conversation/src/client/locales.ts b/packages/client/ui-conversation/src/client/locales.ts index 9b7d24480f..39a013b3e2 100644 --- a/packages/client/ui-conversation/src/client/locales.ts +++ b/packages/client/ui-conversation/src/client/locales.ts @@ -23,6 +23,18 @@ export const zh = { 'input.stop': '停止生成', 'input.send': '发送消息', 'input.accessMode': '访问模式，当前：{name}', + 'context.aria': '上下文已用 {percent}', + 'context.used': '上下文已用', + 'context.system': '系统提示词', + 'context.tools': '工具', + 'context.messages': '对话消息', + 'stats.counts': '{turns} 轮 · {steps} 步', + 'stats.llm': 'LLM {duration}', + 'stats.toolCall': '工具调用 {duration}', + 'stats.ttftAverage': '首 token 平均 {duration}', + 'stats.tokensPerSecond': '{throughput} tok/s', + 'stats.cacheHit': '缓存命中 {percent}%', + 'stats.tokens': '输入 {input} tok · 输出 {output} tok', 'settings.enter.title': '繁忙时 Enter 键行为', 'settings.enter.description': '仅在智能体运行时生效；Cmd/Ctrl+Enter 使用另一行为', 'settings.enter.queue': '排队发送', @@ -79,6 +91,8 @@ export const zh = { 'message.retry.failure': '失败原因：', 'message.turnError': '本轮运行失败', 'message.ranFor': '用时 {duration}', + 'message.ttft': '首 token {seconds}秒', + 'message.tokensPerSecond': '{tps} tok/s', 'duration.seconds': '{seconds}秒', 'duration.minutes': '{minutes}分{seconds}秒', 'command.running': '执行中…', @@ -144,6 +158,18 @@ export const en = { 'input.stop': 'Stop generating', 'input.send': 'Send message', 'input.accessMode': 'Access mode, current: {name}', + 'context.aria': '{percent} of context used', + 'context.used': 'of context used', + 'context.system': 'System prompt', + 'context.tools': 'Tools', + 'context.messages': 'Messages', + 'stats.counts': '{turns} turns · {steps} steps', + 'stats.llm': 'LLM {duration}', + 'stats.toolCall': 'Tool call {duration}', + 'stats.ttftAverage': 'TTFT avg {duration}', + 'stats.tokensPerSecond': '{throughput} tok/s', + 'stats.cacheHit': 'Cache hit {percent}%', + 'stats.tokens': 'Input {input} tok · Output {output} tok', 'settings.enter.title': 'Enter behavior while busy', 'settings.enter.description': 'Busy only; Cmd/Ctrl+Enter uses the other behavior', 'settings.enter.queue': 'Queue', @@ -200,6 +226,8 @@ export const en = { 'message.retry.failure': 'Failure reason: ', 'message.turnError': 'This turn failed', 'message.ranFor': 'Ran for {duration}', + 'message.ttft': 'TTFT {seconds}s', + 'message.tokensPerSecond': '{tps} tok/s', 'duration.seconds': '{seconds}s', 'duration.minutes': '{minutes}m {seconds}s', 'command.running': 'Running…', diff --git a/packages/client/ui-conversation/src/client/skeleton/ContextMeter.module.css b/packages/client/ui-conversation/src/client/skeleton/ContextMeter.module.css new file mode 100644 index 0000000000..9eed743214 --- /dev/null +++ b/packages/client/ui-conversation/src/client/skeleton/ContextMeter.module.css @@ -0,0 +1,147 @@ +/* Context-occupancy ring beside the send button plus its click-open breakdown + panel (menu surface: r12, inverted hairline, shadow-lv3). */ + +.root { + position: relative; + display: inline-flex; +} + +/* Same 28px circular hit target family as the composer's attach button. */ +.trigger { + display: grid; + place-items: center; + flex: none; + width: 28px; + height: 28px; + border: none; + border-radius: 999px; + background: transparent; + color: var(--dsw-alias-label-secondary); + cursor: pointer; +} + +.trigger:hover { + background: var(--dsw-alias-interactive-bg-hover); +} + +.track { + fill: none; + stroke: var(--dsw-alias-border-l3); + stroke-width: 2; +} + +.fill { + fill: none; + stroke: var(--dsw-alias-label-tertiary); + stroke-width: 2; + stroke-linecap: round; +} + +.panel { + position: absolute; + bottom: calc(100% + 8px); + right: 0; + z-index: 100; + box-sizing: border-box; + width: 264px; + padding: 12px; + border: 1px solid var(--dsw-alias-border-inverted); + border-radius: 12px; + background: var(--dsw-specific-menu); + box-shadow: var(--dsw-shadow-lv3); + font-size: 12px; + line-height: 20px; + color: var(--dsw-alias-label-secondary); + cursor: default; +} + +.header { + display: flex; + align-items: center; + gap: 6px; +} + +.figures { + margin-left: auto; + font-weight: 500; + font-variant-numeric: tabular-nums; + color: var(--dsw-alias-label-primary); +} + +.percent { + font-weight: 500; + color: var(--dsw-alias-label-primary); +} + +.headline { + color: var(--dsw-alias-label-tertiary); +} + +/* The headline brackets the reading, so the side a locale leaves empty must + drop out of the flex row rather than spend a gap. */ +.headline:empty { + display: none; +} + +.bar { + display: flex; + gap: 1px; + margin: 10px 0 12px; + height: 4px; + border-radius: 999px; + background: var(--dsw-alias-interactive-bg-hover); + overflow: hidden; +} + +.segment { + flex: none; + min-width: 2px; + height: 100%; + border-radius: 1px; + background: var(--meter-tint, var(--dsw-alias-label-tertiary)); +} + +.swatch { + display: inline-block; + margin-right: 6px; + width: 8px; + height: 8px; + border-radius: 2px; + background: var(--meter-tint); + vertical-align: baseline; +} + +.colorSystem { + --meter-tint: var(--dsw-static-neutral-bluish-400); +} + +.colorTools { + /* The design platform ships no purple static token; violet-400 literal. */ + --meter-tint: rgb(167, 139, 250); +} + +.colorMessages { + --meter-tint: var(--dsw-static-blue-450); +} + +.rows { + margin: 6px 0 0; +} + +.row { + display: flex; + align-items: center; + justify-content: space-between; + gap: 12px; + padding: 2px 0; +} + +.row dt { + color: var(--dsw-alias-label-secondary); +} + +.row dd { + margin: 0; + font-variant-numeric: tabular-nums; + color: var(--dsw-alias-label-primary); +} diff --git a/packages/client/ui-conversation/src/client/skeleton/ContextMeter.tsx b/packages/client/ui-conversation/src/client/skeleton/ContextMeter.tsx new file mode 100644 index 0000000000..1f0bedded7 --- /dev/null +++ b/packages/client/ui-conversation/src/client/skeleton/ContextMeter.tsx @@ -0,0 +1,153 @@ +/** Composer context-occupancy meter: a ring beside the send button fed by the + * `contextPressure` projection, with a click-open panel of the heuristic + * `contextBreakdown` composition (system prompt, tools, conversation). + * Renders nothing until a provider reports both pressure and a route capacity + * (same gate as the stats row used). */ + +import { useEffect, useRef, useState } from 'react' +import type { UseProjection } from '@deepseek-ai/dsh-client-runtime/client' +// Type-only: the `contextPressure` / `contextBreakdown` projection key merges. +import type {} from '@deepseek-ai/dsh-token-meter/client' +import { Tooltip } from '@deepseek-ai/dsh-client-ui-primitives' +import type { ComposerBarProps } from '../contract/slots.ts' +import { contextOccupancy, formatTokens } from '../chat/StatsLine.tsx' +import css from './ContextMeter.module.css' + +/** Ring geometry: 14px viewBox, 2px stroke. */ +const RADIUS = 5.5 +const CIRCUMFERENCE = 2 * Math.PI * RADIUS + +/** + * Marker the localized occupancy sentence is split on, so the panel headline + * keeps the reading in its own tone while each locale still owns the word + * order (`45% of context used` / `上下文已用 45%`). + */ +const READING_SLOT = '\u0000' + +/** Panel legend rows, in bar-segment order; each color class carries the shared swatch/segment tint. */ +const ROWS = [ + { key: 'systemTokens', label: 'context.system', color: css.colorSystem }, + { key: 'toolsTokens', label: 'context.tools', color: css.colorTools }, + { key: 'messageTokens', label: 'context.messages', color: css.colorMessages }, +] as const + +export interface ContextMeterProps { + useProjection: UseProjection + /** The owning bar's locale seat, passed down as a plain prop. */ + t: ComposerBarProps['t'] +} + +export function ContextMeter({ useProjection, t }: ContextMeterProps) { + const pressure = useProjection('contextPressure') + const breakdown = useProjection('contextBreakdown') + const [open, setOpen] = useState(false) + const rootRef = useRef(null) + const context = contextOccupancy(pressure) + const available = context !== null + + // A model switch can temporarily remove capacity while this component stays + // mounted. Close the now-unavailable panel instead of preserving stale UI. + useEffect(() => { + if (!available && open) setOpen(false) + }, [available, open]) + + // Outside click / Escape close, one document listener while open (Menu's pattern). + useEffect(() => { + if (!open || !available) return + const onPointerDown = (e: PointerEvent): void => { + if (e.target instanceof Node && rootRef.current?.contains(e.target) === true) return + setOpen(false) + } + const onKeyDown = (e: KeyboardEvent): void => { + if (e.key === 'Escape') setOpen(false) + } + document.addEventListener('pointerdown', onPointerDown) + document.addEventListener('keydown', onKeyDown) + return () => { + document.removeEventListener('pointerdown', onPointerDown) + document.removeEventListener('keydown', onKeyDown) + } + }, [available, open]) + + if (context === null) return null + const percent = context.percent + const reading = `${percent}%` + const [headBefore = '', headAfter = ''] = t('context.aria', { percent: READING_SLOT }) + .split(READING_SLOT) + .map(part => part.trim()) + + // The bar's overall length stays the provider-exact percent; the heuristic + // breakdown only proportions its colored parts. A zero-width part is dropped + // instead of rendered: `.segment`'s min-width keeps a hairline part visible, + // which at 0% occupancy would draw a filled bar over an empty context. + const breakdownTotal = breakdown === undefined + ? 0 + : breakdown.systemTokens + breakdown.toolsTokens + breakdown.messageTokens + const parts = breakdown === undefined || breakdownTotal === 0 + ? [{ key: 'total', color: undefined, width: percent }] + : ROWS.map(row => ({ key: row.key, color: row.color, width: percent * breakdown[row.key] / breakdownTotal })) + const segments = parts.filter(part => part.width > 0) + + return ( + + + + + {open && ( +
+
+ {/* Empty sides collapse through `.headline:empty` so the locale that + needs no leading (or trailing) text spends no header gap. */} + {headBefore} + {reading} + {headAfter} + + {`~${formatTokens(context.usedTokens)} / ${formatTokens(context.contextWindow)}`} + +
+
+ {segments.map(segment => ( +
+ ))} +
+ {breakdown !== undefined && ( +
+ {ROWS.map(row => ( +
+
+ + {t(row.label)} +
+
{`~${formatTokens(breakdown[row.key])}`}
+
+ ))} +
+ )} +
+ )} + + ) +} diff --git a/packages/client/ui-conversation/src/client/skeleton/InputBar.tsx b/packages/client/ui-conversation/src/client/skeleton/InputBar.tsx index 5f09e7315a..131f63c49d 100644 --- a/packages/client/ui-conversation/src/client/skeleton/InputBar.tsx +++ b/packages/client/ui-conversation/src/client/skeleton/InputBar.tsx @@ -19,6 +19,7 @@ import type { Translate } from '@deepseek-ai/dsh-client-ui-slots' import type { ComposerBarProps } from '../contract/slots.ts' import { deriveDecorations } from '../input/decorations.ts' import type { DraftDecorations } from '../input/decorations.ts' +import { ContextMeter } from './ContextMeter.tsx' import { PermissionSelect } from './PermissionSelect.tsx' import css from './InputBar.module.css' @@ -512,6 +513,7 @@ export function InputBar({
{rightItems} {renderSlot('conversation.input.model', { locked })} + {/* {machineBusy && } */}