Coder/deepseek-harness
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(replay): mark local compaction calls

Browse Source
This commit is contained in:
Tianyi Cui
2026-08-08 02:16:17 +08:00
parent d301dbdc7d
commit 8dcfe5c406
29 changed files with 147 additions and 50 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
.agents/notes/implemented/bug-fix/2026-07-21-compaction-summary-prefix-cache-reuse.i18n.yaml
View File

@@ -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 .agents/notes/implemented/bug-fix/2026-07-21-compaction-summary-prefix-cache-reuse.md
2026-07-21-compaction-summary-prefix-cache-reuse.md: d05d25cfa7c3984ce0ce75c38068a91a0e07dfe8
2026-07-21-compaction-summary-prefix-cache-reuse.zh.md: f31c33e680b3db60a6fb758a522addca63224fea
2026-07-21-compaction-summary-prefix-cache-reuse.md: fb6524846f837b783423b3ed0e043cd4e1f7849d
2026-07-21-compaction-summary-prefix-cache-reuse.zh.md: e6072313b7089b9dda14c298fcb94db786220501

4
.agents/notes/implemented/bug-fix/2026-07-21-compaction-summary-prefix-cache-reuse.md
View File

@@ -29,7 +29,7 @@ Auto-compaction always anchors at the surface head, so the shadowed region is th
- **Keep the summarizer system prompt but reuse the rest** — rejected: the system slot is the very first token region a provider caches on, so a distinct summarizer system prompt invalidates the whole prefix regardless of what follows. Only moving the directive off the front recovers the cache.
- **Send only the shadowed region without the `system`/`tools` head** — rejected: a differently-headed sequence still diverges from the cached request at the first token, so it caches no better while losing the framing the summary needs.
- **Omit `tools` from the summarization request** (the model never calls one) — rejected: tool schemas are part of the cached token sequence; omitting them misaligns every following token and defeats reuse.
- **A dedicated `assistant/chunk`-emitting summarization sub-session for snapshot replay** — out of scope here; the replay gap predates this change and is tracked in the [compaction-seam note](../feature/2026-06-18-compaction-capability-seam.md).
- **A dedicated `assistant/chunk`-emitting summarization sub-session for snapshot replay** — rejected: the durable `compact/summary` event records the successful local call's position and complete output, while its explicit call marker prevents replay from treating template or remote output as a local stream.
## Consequences
@@ -42,4 +42,4 @@ Auto-compaction always anchors at the surface head, so the shadowed region is th
- **Unit:** `compact-basic.spec.ts` asserts the auxiliary call forwards `system`/`tools`/leading messages and appends the compaction instruction as the final message, and that `compactRegion` replays the latest routed header prefix. Existing content assertions read the summarizer input through the replayed messages rather than a transcript string.
- **Loop:** `compact-loop-repro.spec.ts` classifies the summarization request by the compaction instruction in its trailing user message, and the overflow-recovery tests continue to pin conversation-vs-summary request counts across the real loop.
- **Snapshot gap unchanged:** the summarization call still emits no `assistant/chunk` events, so it remains outside keyless replay; the pre-existing gap is owned by the [compaction-seam note](../feature/2026-06-18-compaction-capability-seam.md).
- **Snapshot:** keyless replay reconstructs one canonical successful stream from a marked `compact/summary`; the [compaction-seam note](../feature/2026-06-18-compaction-capability-seam.md) owns the durable marker contract.

4
.agents/notes/implemented/bug-fix/2026-07-21-compaction-summary-prefix-cache-reuse.zh.md
View File

@@ -29,7 +29,7 @@ Status: implemented
- **保留摘要器系统提示词但复用其余部分**——否决：system 槽位正是提供方最先做缓存的 token 区域，因此一个不同的摘要器系统提示词无论后面跟着什么都会使整个前缀失效。只有把指令移离前端才能恢复缓存。
- **只发送被遮蔽区域而不带 `system`/`tools` 头部**——否决：头部不同的序列在第一个 token 处仍然与已缓存请求分叉，因此缓存效果并不更好，反而丢失了摘要所需的框架。
- **从摘要请求中省略 `tools`**（模型从不调用任何工具）——否决：工具 schema 是已缓存 token 序列的一部分；省略它们会让后续每个 token 失去对齐，破坏复用。
- **为快照回放专门建立一个发出 `assistant/chunk` 的摘要子会话**——此处超出范围；该回放缺口早于本次改动，记录在 [compaction-seam Agent Note](../feature/2026-06-18-compaction-capability-seam.md) 中。
- **为快照回放专门建立一个发出 `assistant/chunk` 的摘要子会话**——否决：持久的 `compact/summary` 事件会记录成功本地调用的位置和完整输出，而显式调用标记可防止回放把模板或远程输出当作本地流。
## 后果
@@ -42,4 +42,4 @@ Status: implemented
- **单元：** `compact-basic.spec.ts` 断言辅助调用转发 `system`/`tools`/前导消息，并把压缩指令作为最后一条消息追加，且 `compactRegion` 回放最新的已路由 header 前缀。现有的内容断言通过回放的消息而非 transcript 字符串来读取摘要器输入。
- **循环：** `compact-loop-repro.spec.ts` 依据摘要请求尾部 user 消息中的压缩指令对其分类，溢出恢复测试则继续在真实循环中固定对话请求与摘要请求的数量。
- **快照缺口未变：** 摘要调用仍然不发出 `assistant/chunk` 事件，因此它仍处于无密钥回放之外；这一既有缺口归 [compaction-seam Agent Note](../feature/2026-06-18-compaction-capability-seam.md) 所有。
- **快照：** 无密钥回放会从带标记的 `compact/summary` 重建一条规范成功流；[compaction-seam Agent Note](../feature/2026-06-18-compaction-capability-seam.md) 负责持久标记契约。

4
.agents/notes/implemented/feature/2026-06-18-compaction-capability-seam.i18n.yaml
View File

@@ -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 .agents/notes/implemented/feature/2026-06-18-compaction-capability-seam.md
2026-06-18-compaction-capability-seam.md: 8dcbe74429a620027a570124383442b969c12196
2026-06-18-compaction-capability-seam.zh.md: 27b63f29c2e6f35637185b47c882ae42e5d41088
2026-06-18-compaction-capability-seam.md: 390d091fd6e6f7fa1694ba78b90522b70948fe92
2026-06-18-compaction-capability-seam.zh.md: d72793e6adde911a8623ac1c6b65df899381923e

6
.agents/notes/implemented/feature/2026-06-18-compaction-capability-seam.md
View File

@@ -33,7 +33,7 @@ This is not a coupling smell — it is the contract's domain. The "only cordis"
An earlier draft put the full algorithm (the retention walk, token-summing, text extraction) as concrete methods on the interface. That recouples the contract to one strategy: a backend that wants a different retention policy or event sequence would have to fight inherited concrete code. Making all three operations abstract puts every *how* decision in the backend and keeps the interface a statement of *what*. Token measurement is not a compaction hook at all; the singleton service lets multiple consumers share one per-session replay fold.
`compactIfNeeded(agent, trigger, signal)` takes an explicit `'pressure' | 'context-overflow'` trigger and cancellation. It reads only the latest durable routed request; no header means no work, while any routed provider/model target uses the singleton estimator. `compactNow(agent, signal)` requires an idle agent and performs one useful balanced reduction even below pressure, returning `null` without writes when none exists. `compactRegion(start, end, agent, signal?)` uses `agent.session` as its single session identity and keeps an optional signal for explicit callers. The default summarizer resolves its target from explicit config, the latest logged routed target, then agent options, and records the provider/model pair after any `llm/stream` routing. It replays the routed request's prefix and appends the compaction directive as a trailing user message so the provider's warm KV cache is reused — see the [summary prefix-cache Agent Note](../bug-fix/2026-07-21-compaction-summary-prefix-cache-reuse.md). The call sets the provider-neutral `GenerateOptions.purpose` to `compaction`; adapters may map that purpose to model-hidden transport metadata, and the DeepSeek adapter sends `x-deepseek-harness-compact: 1`.
`compactIfNeeded(agent, trigger, signal)` takes an explicit `'pressure' | 'context-overflow'` trigger and cancellation. It reads only the latest durable routed request; no header means no work, while any routed provider/model target uses the singleton estimator. `compactNow(agent, signal)` requires an idle agent and performs one useful balanced reduction even below pressure, returning `null` without writes when none exists. `compactRegion(start, end, agent, signal?)` uses `agent.session` as its single session identity and keeps an optional signal for explicit callers. The default summarizer resolves its target from explicit config, the latest logged routed target, then agent options, and records the provider/model pair after any `llm/stream` routing. It replays the routed request's prefix and appends the compaction directive as a trailing user message so the provider's warm KV cache is reused — see the [summary prefix-cache Agent Note](../bug-fix/2026-07-21-compaction-summary-prefix-cache-reuse.md). The result carries `llmStreamCall: true` because it consumed exactly one call through this context's LLM service; a subclass sets that marker only under the same condition, since retained `rawOutput` alone does not identify the call path. The call sets the provider-neutral `GenerateOptions.purpose` to `compaction`; adapters may map that purpose to model-hidden transport metadata, and the DeepSeek adapter sends `x-deepseek-harness-compact: 1`.
### Automatic pressure runs after successful durable step work
@@ -76,7 +76,7 @@ Because `SurfaceEventType` is closed, the summary cannot ride on a `compact/*` e
```
compact/start → log-only. Acquires the lock.
[summarize older range via the backend]
compact/summary → log-only. Provenance: raw summary, range, shadowed seqs, token count.
compact/summary → log-only. Provenance: raw summary, local-call marker, range, shadowed seqs, token count.
user/message → canonical checkpoint source + surfaceOp { op:'replace', start, end }.
THE surface mutation (framed summary).
deriveMessages() renders it as a user-role message.
@@ -131,4 +131,4 @@ The lifecycle boundary makes crash state unambiguous:
- **Loop:** Tests pin pre-step after the preceding `step/end` and before the next `step/start`, actual `agent/request` routing, closed failed steps, fresh retry numbering, and complete thrown/in-band overflow → compaction → reconstructed retry composition.
- **Manual:** Maintenance serialization, marker ordering, injection retention, live/stale orphan classification, cancellation, close/flush failures, command mapping, and the queued TUI journey are pinned without a model key.
- **With-key e2e:** A real model and bash session with lowered limits triggers compaction, records a complete `compact/start…end` pair, shrinks the surface, and finishes the task.
- **Snapshot gap:** The summarization call is session-associated and logs `compact/summary`, but ordinary transcript replay does not derive its auxiliary response; keyless assembled coverage therefore needs an explicit replay override.
- **Snapshot:** The assembled context-overflow scenario derives the auxiliary call from `compact/summary` only when `llmStreamCall: true` proves that the local LLM service consumed it; canonical reconstructed blocks pin the complete recovery without provider delta partitioning.

6
.agents/notes/implemented/feature/2026-06-18-compaction-capability-seam.zh.md
View File

@@ -33,7 +33,7 @@ Status: implemented
早期草案将完整算法（保留遍历、token 求和、文本提取）作为接口上的具体方法。这会将契约重新耦合到一种策略：想要不同保留策略或事件排序的后端必须与继承来的具体代码对抗。将三个操作都设为抽象，把所有*怎么做*的决策放在后端，并让接口保持为*做什么*的声明。token 测量根本不是压缩钩子；单例服务使多个消费方能够共享逐会话的回放折叠。
`compactIfNeeded(agent, trigger, signal)` 接受显式的 `'pressure' | 'context-overflow'` 触发原因与取消信号。它只读取最新的持久化已路由请求；没有 header 就不执行工作，任何已路由的提供方/模型目标都使用单例估算器。`compactNow(agent, signal)` 要求 agent 处于 idle，即使未达到压力也进行一次有效的平衡缩减；不存在这种范围时返回 `null`，且不写入任何内容。`compactRegion(start, end, agent, signal?)` 将 `agent.session` 作为唯一会话身份，并为显式调用方保留可选 signal。默认摘要器依次从显式配置、最新记录的已路由目标和 agent 选项解析目标，并在任何 `llm/stream` 路由后记录提供方/模型对。它回放已路由请求的前缀，并将压缩指令追加为尾部 user 消息，从而复用提供方的热 KV cache；见[摘要前缀缓存 Agent Note](../bug-fix/2026-07-21-compaction-summary-prefix-cache-reuse.md)。该调用将提供方无关的 `GenerateOptions.purpose` 设为 `compaction`；适配器可以将此用途映射为对模型隐藏的传输元数据，DeepSeek 适配器会发送 `x-deepseek-harness-compact: 1`。
`compactIfNeeded(agent, trigger, signal)` 接受显式的 `'pressure' | 'context-overflow'` 触发原因与取消信号。它只读取最新的持久化已路由请求；没有 header 就不执行工作，任何已路由的提供方/模型目标都使用单例估算器。`compactNow(agent, signal)` 要求 agent 处于 idle，即使未达到压力也进行一次有效的平衡缩减；不存在这种范围时返回 `null`，且不写入任何内容。`compactRegion(start, end, agent, signal?)` 将 `agent.session` 作为唯一会话身份，并为显式调用方保留可选 signal。默认摘要器依次从显式配置、最新记录的已路由目标和 agent 选项解析目标，并在任何 `llm/stream` 路由后记录提供方/模型对。它回放已路由请求的前缀，并将压缩指令追加为尾部 user 消息，从而复用提供方的热 KV cache；见[摘要前缀缓存 Agent Note](../bug-fix/2026-07-21-compaction-summary-prefix-cache-reuse.md)。该结果携带 `llmStreamCall: true`，因为生成它时恰好通过此上下文的 LLM 服务发起了一次调用；只有满足相同条件时，子类才设置该标记，因为单有保留的 `rawOutput` 并不能判定调用路径。该调用将提供方无关的 `GenerateOptions.purpose` 设为 `compaction`；适配器可以将此用途映射为对模型隐藏的传输元数据，DeepSeek 适配器会发送 `x-deepseek-harness-compact: 1`。
### 成功的持久步骤工作完成后运行自动压力检查
@@ -76,7 +76,7 @@ retry → next numbered step/start ⟵ derives from the replacement surface
```
compact/start → log-only. Acquires the lock.
[summarize older range via the backend]
compact/summary → log-only. Provenance: raw summary, range, shadowed seqs, token count.
compact/summary → log-only. Provenance: raw summary, local-call marker, range, shadowed seqs, token count.
user/message → canonical checkpoint source + surfaceOp { op:'replace', start, end }.
THE surface mutation (framed summary).
deriveMessages() renders it as a user-role message.
@@ -131,4 +131,4 @@ compact/end → log-only. Releases the lock (carries `error` on a recoverab
- **循环测试：** 测试固定 pre-step 发生在前一个 `step/end` 之后、下一个 `step/start` 之前，使用实际 `agent/request` 路由，关闭失败步骤，分配新的重试编号，并覆盖完整的抛出/带内溢出 → 压缩 → 重建重试组合。
- **手动测试：** 无需模型密钥即可固定 maintenance 串行化、标记顺序、注入保留、活动／陈旧未匹配标记分类、取消、闭合／flush 失败、命令映射以及排队 TUI 流程。
- **带密钥 e2e：** 真实模型和 bash 会话在降低的限制下触发压缩，记录完整的 `compact/start…end` 对，缩小 surface，并完成任务。
- **快照缺口：** 摘要调用与会话关联并记录 `compact/summary`，但普通 transcript（文本记录）回放不会派生其辅助响应；因此，要实现无密钥的组装态覆盖，就必须显式提供回放 override。
- **快照：** 组装后的上下文溢出场景仅在 `llmStreamCall: true` 证明本地 LLM 服务执行了辅助调用时，才从 `compact/summary` 派生该调用；规范重建的块在不固定提供方增量切分的情况下固定完整恢复过程。

4
.agents/notes/implemented/testing/2026-06-19-acp-snapshot-tests.i18n.yaml
View File

@@ -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 .agents/notes/implemented/testing/2026-06-19-acp-snapshot-tests.md
2026-06-19-acp-snapshot-tests.md: c7b95bd68027705b99d850d596405e56eea0dfca
2026-06-19-acp-snapshot-tests.zh.md: 43d43262684920cae5feedb5f2eb109db00f9f8c
2026-06-19-acp-snapshot-tests.md: c6e81dc4a51083ed37ad6f837fe7dbb08c212e23
2026-06-19-acp-snapshot-tests.zh.md: 5a579c3373442dc4cfc7d661ea2f2f398b804dc7

7
.agents/notes/implemented/testing/2026-06-19-acp-snapshot-tests.md
View File

@@ -24,7 +24,7 @@ Every committed session-format fixture uses the canonical packed physical layout
### Replay derives the model script from the log
`llm-replay` short-circuits the provider-agnostic `llm/stream` waterfall. `deriveReplayScript()` groups recorded chunks by `(turn, step)` and serves one group per model call. The loop makes one stream call per step, so the grouping is exact and includes error finish chunks without special handling.
`llm-replay` short-circuits the provider-agnostic `llm/stream` waterfall. `deriveReplayScript()` splits recorded `assistant/chunk` events at terminal `finish` chunks and uses `(turn, step)` changes to reject an unterminated prior call. A `compact/summary` with `llmStreamCall: true` contributes one call at its durable log position: replay reconstructs canonical block boundaries from `rawOutput`, retains recorded usage when present, and supplies a terminal `stop`. The marker distinguishes that local call from template or remote summaries whose retained `rawOutput` did not consume this context's adapter.
### The in-memory replay entry honors the full LLM contract
@@ -36,7 +36,7 @@ Every committed session-format fixture uses the canonical packed physical layout
| { kind: 'hang' }
```
Logs derive chunk entries. Pre-stream throws and hangs have no reconstructable chunk representation, so those scenarios provide `replay.override.json`. A throw entry may include prefix chunks for mid-stream failure. Explicit overrides avoid inferring adapter behavior from lossy turn-end reasons.
Logs derive chunk entries from finished assistant streams and explicitly marked compaction calls. Pre-stream throws, hangs, and external summarizer calls have no reconstructable local chunk representation, so those scenarios provide `replay.override.json`. A throw entry may include prefix chunks for mid-stream failure. Explicit overrides avoid inferring adapter behavior from lossy turn-end reasons or provider output alone.
### Positional replay, one in-flight stream
@@ -74,12 +74,13 @@ Tool determinism comes from a generated cwd, scrubbed environment, fresh non-log
## Alternatives considered
- **A hand-authored `llm.json` of model chunks** — the earlier draft; reusing the real session log makes the fixture a genuine product of the system rather than a hand-built mock, and doubles it as a behavioral expected output.
- **A compulsory replay override for every compaction summary** — rejected: the durable summary event already fixes a successful local call's position, complete output, and optional usage. An explicit local-call marker preserves that single-source fixture without inventing a call for template or remote summarizers.
- **A byte-level HTTP-record library (Polly/nock/MSW)** — rejected: adapter-specific, awkward with streaming SSE, and lower-level than the thing under test.
- **Synthesizing throw/cancel entries from `turn/end {kind:'error'|'aborted'}`** — rejected: it couples `llm-replay` to loop-internal turn-closing semantics, and the `turn/end` reason is lossy (it cannot distinguish a thrown 401 from a finish-error); the explicit `replay.override.json` sidecar is the cleaner seam.
- **Copying both request-header sidecars beside every class pin** — rejected: prompt and tool-schema composition vary independently, so a change to one shared component would churn byte-identical files across unrelated class pins. Explicit per-component sources retain one structural pin per class without duplicating content.
## Consequences
The tier adds reviewed per-scenario input, session, stdout, optional override, and optional workspace fixtures, plus one file for each distinct pinned prompt and tool-schema sequence. Workspace seeds are copied into the generated cwd for both record and replay. In return the tier provides deterministic keyless coverage through the real Loader and tool composition. Most retained scenarios exercise the assembled backend rather than ACP; the [automation-only ACP decision](../simplification/2026-07-23-acp-automation-only-protocol.md#snapshot-boundary) keeps that corpus here until it can move to a transport-neutral headless suite without losing coverage.
The tier adds reviewed per-scenario input, session, stdout, optional override, and optional workspace fixtures, plus one file for each distinct pinned prompt and tool-schema sequence. Workspace seeds are copied into the generated cwd for both record and replay. In return the tier provides deterministic keyless coverage through the real Loader and tool composition, including an assembled context-overflow recovery whose marked compaction summary supplies the auxiliary call. Most retained scenarios exercise the assembled backend rather than ACP; the [automation-only ACP decision](../simplification/2026-07-23-acp-automation-only-protocol.md#snapshot-boundary) keeps that corpus here until it can move to a transport-neutral headless suite without losing coverage.
This Agent Note relates to but does not supersede the [proposed determinism Agent Note](../../proposed/testing/2026-06-11-deterministic-and-stress-testing.md): that proposal's "universal replay fixture" re-derives session *message history* after every test (an internal-consistency invariant), whereas these snapshots pin assembled behavior plus the external automation output. They are complementary until the backend corpus moves off ACP.

7
.agents/notes/implemented/testing/2026-06-19-acp-snapshot-tests.zh.md
View File

@@ -24,7 +24,7 @@ Status: implemented
### 回放从日志推导模型脚本
`llm-replay` 短路了提供方无关的 `llm/stream` waterfall（瀑布式事件）。`deriveReplayScript()` 按 `(turn, step)` 对已录制的分片分组，每次模型调用服务一组。agent loop（智能体循环）每个步骤发起一次流调用，因此分组精确对应，错误结束分片也无需特殊处理。
`llm-replay` 短路了提供方无关的 `llm/stream` waterfall（瀑布式事件）。`deriveReplayScript()` 在终止的 `finish` 分片处切分已记录的 `assistant/chunk` 事件，并用 `(turn, step)` 变化拒绝前一条未终止的调用。携带 `llmStreamCall: true` 的 `compact/summary` 会在其持久日志位置贡献一次调用：回放根据 `rawOutput` 重建规范块边界，保留已记录的 usage（如有），并提供终止的 `stop`。该标记将这次本地调用与模板摘要或远程摘要区分开；后两者即使保留了 `rawOutput`，也未使用此上下文的适配器。
### 内存中的回放条目遵守完整的 LLM 契约
@@ -36,7 +36,7 @@ Status: implemented
| { kind: 'hang' }
```
日志推导出分片条目。流开始前的抛出和挂起没有可重建的分片表示，因此这些场景提供 `replay.override.json`。throw 条目可以包含前缀分片以模拟流中途失败。显式覆盖避免了从有损的轮次结束原因推断适配器行为。
日志从已结束的 assistant 流和显式标记的压缩（compaction）调用推导分片条目。流开始前的抛出、挂起和外部摘要器调用没有可重建的本地分片表示，因此这些场景提供 `replay.override.json`。throw 条目可以包含前缀分片以模拟流中途失败。显式覆盖避免了从有损的轮次结束原因或单独的提供方输出推断适配器行为。
### 位置式回放，单个在途流
@@ -74,12 +74,13 @@ Status: implemented
## 曾考虑的替代方案
- **手工编写包含模型分片的 `llm.json`**——早期草案；复用真实会话日志，使 fixture 成为系统的真实产物而非手工构建的 mock，并让它同时充当行为预期输出。
- **为每个压缩摘要强制提供回放 override**——否决：持久摘要事件已经固定成功本地调用的位置、完整输出与可选 usage。显式的本地调用标记保留了这份单一来源 fixture，而不会为模板摘要器或远程摘要器凭空构造调用。
- **字节级 HTTP 录制库（Polly/nock/MSW）**：否决。与适配器耦合，处理流式 SSE（Server-Sent Events）时笨拙，且层级低于被测对象。
- **从 `turn/end {kind:'error'|'aborted'}` 合成抛错/取消条目**：否决。这会将 `llm-replay` 耦合到 loop 内部的轮次关闭语义，且 `turn/end` 原因是有损的（无法区分抛出的 401 与 finish-error）；显式的 `replay.override.json` 伴随文件是更清晰的 seam。
- **在每个类别 pin 旁复制两个请求头伴随文件**：否决。提示词与工具 schema 的组合各自独立变化，因此一个共享组件发生变更，就会使不相关类别 pin 中字节完全相同的文件产生无意义改动。显式的分组件来源可在不重复内容的情况下，为每个类别保留一个结构性 pin。
## 后果
该测试层为每个场景增加经过评审的输入、会话、stdout、可选 override 和可选 workspace fixture，并为每个不同的已固定提示词序列、每个不同的已固定工具 schema 序列各增加一个文件。记录与回放都会把 workspace seed 复制到生成的 cwd。作为回报，该层通过真实 Loader 和工具组合提供确定性的无密钥覆盖。保留下来的大多数场景测试的是组装后的后端而非 ACP；[仅面向自动化的 ACP 决策](../simplification/2026-07-23-acp-automation-only-protocol.md#snapshot-boundary)将该语料保留在此处，直至它能够在不损失覆盖的情况下迁移到传输无关的 headless 套件。
该测试层为每个场景增加经过评审的输入、会话、stdout、可选 override 和可选 workspace fixture，并为每个不同的已固定提示词序列、每个不同的已固定工具 schema 序列各增加一个文件。记录与回放都会把 workspace seed 复制到生成的 cwd。作为回报，该层通过真实 Loader 和工具组合提供确定性的无密钥覆盖，其中包括一个组装后的上下文溢出恢复场景，其带标记的压缩摘要提供辅助调用。保留下来的大多数场景测试的是组装后的后端而非 ACP；[仅面向自动化的 ACP 决策](../simplification/2026-07-23-acp-automation-only-protocol.md#snapshot-boundary)将该语料保留在此处，直至它能够在不损失覆盖的情况下迁移到传输无关的 headless 套件。
本 Agent Note 与[拟议的确定性 Agent Note](../../proposed/testing/2026-06-11-deterministic-and-stress-testing.md)相关，但不取代它：该提案的“通用回放 fixture”在每次测试后重新派生会话*消息历史*（内部一致性不变量），而这些快照固定组装后的行为与外部自动化输出。在后端语料迁出 ACP 之前，两者相互补充。