refactor(agent): complete inbox lifecycle migration

This commit is contained in:
_Kerman
2026-08-03 12:25:33 +08:00
parent dc1d542092
commit 49e90695cc
214 files changed
+6019 -4235

No files matched your search

@@ -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/compaction.md
compaction.md: 0ccd38b2f70bce422d064acfd33cbbb93c088af2
compaction.zh.md: 3fc21944e836372a621b9934f237e1f05d5f18de
compaction.md: fe64ffe2707ab4a41f1db186965b944d525684c3
compaction.zh.md: 7b4f8e5e9aa57cc575b46f79ce7a6c44314a21c4
+8 -2
View File
@@ -62,13 +62,19 @@ Automatic callers state why policy is running; implementations may treat confirm
type CompactionTrigger = 'pressure' | 'context-overflow'
```
`CompactService` exposes `compactIfNeeded(agent, trigger, signal)` for automatic `pressure` or `context-overflow` policy, `compactNow(agent, signal)` for one useful idle-session reduction even below pressure, and `compactRegion(...)` for an explicit inclusive surface range. `compactNow()` synchronously reserves the agent's next-turn admission, returns `null` without writing when no useful range exists, records a standalone `turn: null` bracket before summarization, flushes a closed attempt, and then releases admission so ordinary queued prompts derive from the new surface. Every backend marks its replacement `user/message` with `COMPACT_CHECKPOINT_SOURCE`; client and wire consumers import that value and `isCompactCheckpointSource()` from the cordis-free `@deepseek-ai/dsh-compact/checkpoint` subpath, while the package root re-exports both for host consumers. The predicate keeps checkpoint recognition independent of any one backend. Implementations must forward the supplied signal to summarization. The seam owns no pricing API: the singleton [`ctx.tokenMeter`](token-meter.md) directly owns estimation and replay, while `dsh-compact-basic` owns retention, event sequencing, routed summarization calls, and their configuration.
`CompactService` exposes `compactIfNeeded(agent, trigger, signal)` for automatic `pressure` or `context-overflow` policy, `compactNow(agent, signal)` for one useful idle-session reduction even below pressure, and `compactRegion(...)` for an explicit inclusive surface range. `compactNow()` runs as agent maintenance between turns, returns `null` without writing when no useful range exists, records a standalone `turn: null` bracket before summarization, and flushes a closed attempt before later queued prompts may derive from the new surface. Every backend marks its replacement `user/message` with `COMPACT_CHECKPOINT_SOURCE`; client and wire consumers import that value and `isCompactCheckpointSource()` from the cordis-free `@deepseek-ai/dsh-compact/checkpoint` subpath, while the package root re-exports both for host consumers. The predicate keeps checkpoint recognition independent of any one backend. Implementations must forward the supplied signal to summarization. The seam owns no pricing API: the singleton [`ctx.tokenMeter`](token-meter.md) directly owns estimation and replay, while `dsh-compact-basic` owns retention, event sequencing, routed summarization calls, and their configuration.
Expected manual failures use `ManualCompactionErrorCode`:
```ts type-equiv
/** Expected failure classes for an explicit idle-session compaction request. */
type ManualCompactionErrorCode = 'busy' | 'changed' | 'summary' | 'commit' | 'persistence'
type ManualCompactionErrorCode =
| 'busy'
| 'cancelled'
| 'changed'
| 'summary'
| 'commit'
| 'persistence'
```
`changed` and `summary` leave the conversation surface unchanged but still close and persist the failed attempt in the log. `commit` may follow partial mutation; `persistence` means the in-memory bracket closed but its flush failed. Cancellation remains separate and throws the exact abort reason after required cleanup.
+8 -2
View File
@@ -62,13 +62,19 @@ interface CompactionResult {
type CompactionTrigger = 'pressure' | 'context-overflow'
```
`CompactService` 暴露 `compactIfNeeded(agent, trigger, signal)` 以执行自动 `pressure` 或 `context-overflow` 策略,暴露 `compactNow(agent, signal)` 以便即使未达到压力也对空闲会话进行一次有效缩减,还针对显式、两端均包含的 surface 范围暴露 `compactRegion(...)`。`compactNow()` 会同步预留 agent 的下一轮次接纳;没有有效范围时返回 `null` 且不写入;在摘要前记录独立的 `turn: null` 标记对;flush 已闭合尝试;随后释放接纳预留,使普通排队提示词从新表层派生。每个后端都使用 `COMPACT_CHECKPOINT_SOURCE` 标记其替换用的 `user/message`client 与 wire 消费方从无 cordis 的 `@deepseek-ai/dsh-compact/checkpoint` 子路径导入该值和 `isCompactCheckpointSource()`,包根则为 host 消费方重新导出两者。该判定函数使检查点识别不依赖任一特定后端。实现必须把传入的 signal 转发给摘要流程。该 seam 不拥有计价 API:单例 [`ctx.tokenMeter`](token-meter.md) 直接拥有估算与回放,而 `dsh-compact-basic` 拥有保留策略、事件排序、按路由执行的摘要调用及其配置。
`CompactService` 暴露 `compactIfNeeded(agent, trigger, signal)` 以执行自动 `pressure` 或 `context-overflow` 策略,暴露 `compactNow(agent, signal)` 以便即使未达到压力也对空闲会话进行一次有效缩减,还针对显式、两端均包含的 surface 范围暴露 `compactRegion(...)`。`compactNow()` 作为轮次之间的 agent maintenance 运行;没有有效范围时返回 `null` 且不写入;在摘要前记录独立的 `turn: null` 标记对,并在后续排队提示词能够从新表层派生前 flush 已闭合尝试。每个后端都使用 `COMPACT_CHECKPOINT_SOURCE` 标记其替换用的 `user/message`client 与 wire 消费方从无 cordis 的 `@deepseek-ai/dsh-compact/checkpoint` 子路径导入该值和 `isCompactCheckpointSource()`,包根则为 host 消费方重新导出两者。该判定函数使检查点识别不依赖任一特定后端。实现必须把传入的 signal 转发给摘要流程。该 seam 不拥有计价 API:单例 [`ctx.tokenMeter`](token-meter.md) 直接拥有估算与回放,而 `dsh-compact-basic` 拥有保留策略、事件排序、按路由执行的摘要调用及其配置。
预期的手动失败使用 `ManualCompactionErrorCode`
```ts type-equiv
/** Expected failure classes for an explicit idle-session compaction request. */
type ManualCompactionErrorCode = 'busy' | 'changed' | 'summary' | 'commit' | 'persistence'
type ManualCompactionErrorCode =
| 'busy'
| 'cancelled'
| 'changed'
| 'summary'
| 'commit'
| 'persistence'
```
`changed` 和 `summary` 保持会话表层不变,但仍会闭合失败尝试并将其持久化到日志。`commit` 可能发生在部分变更之后;`persistence` 表示内存中的标记对已闭合,但 flush 失败。取消独立于这些失败,并在完成必要清理后抛出原始 abort 原因。
+2 -2
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 docs/core-data-structures/core.md
core.md: 6bb01c91446fbb51e3c472bba3ed3a51b56704ff
core.zh.md: c52bf426020c591aab66d66658b46dfd8dc9b8af
core.md: 32ba11131c660b29a6059006ded841600840f558
core.zh.md: e4fd7440ca240c3104b9b86d6aa516d83b3c452f
+14 -3
View File
@@ -528,9 +528,9 @@ interface Agent {
/**
* Clear queued and steering work — unless `keepInbox` — and abort the active
* turn. The first cause wins for the active turn. Idle cancellation is a
* no-op and does not arm later work.
* @param cause - the stable caller intent carried by the current turn signal.
* turn or between-turn task. The first cause wins for that activity. With no
* active activity, cancellation is a no-op and does not arm later work.
* @param cause - the stable caller intent carried by the active operation signal.
* @param options - cancellation options; `keepInbox` preserves pending work.
*/
cancel(cause: AgentCancelCause, options?: CancelOptions): void
@@ -543,6 +543,17 @@ interface Agent {
*/
whenIdle(): Promise<void>
/**
* Run one non-turn maintenance task from the true idle phase. The task starts
* synchronously after claiming that phase; later waking input remains in the
* inbox until the task settles, while public status stays `idle`.
* `whenIdle()` follows both the task and any waking work released behind it.
* @param task - operation whose fulfillment or rejection is preserved, with a signal aborted by {@link cancel}.
* @throws synchronously when turn-driving or another maintenance task already owns the agent.
* @returns the task promise.
*/
runMaintenance<T>(task: (signal: AbortSignal) => Promise<T>): Promise<T>
/**
* Route identified input to an inbox boundary and optionally wake the driver.
* Waking input submitted after active cancellation is queued for the next turn.
+14 -3
View File
@@ -536,9 +536,9 @@ interface Agent {
/**
* Clear queued and steering work — unless `keepInbox` — and abort the active
* turn. The first cause wins for the active turn. Idle cancellation is a
* no-op and does not arm later work.
* @param cause - the stable caller intent carried by the current turn signal.
* turn or between-turn task. The first cause wins for that activity. With no
* active activity, cancellation is a no-op and does not arm later work.
* @param cause - the stable caller intent carried by the active operation signal.
* @param options - cancellation options; `keepInbox` preserves pending work.
*/
cancel(cause: AgentCancelCause, options?: CancelOptions): void
@@ -551,6 +551,17 @@ interface Agent {
*/
whenIdle(): Promise<void>
/**
* Run one non-turn maintenance task from the true idle phase. The task starts
* synchronously after claiming that phase; later waking input remains in the
* inbox until the task settles, while public status stays `idle`.
* `whenIdle()` follows both the task and any waking work released behind it.
* @param task - operation whose fulfillment or rejection is preserved, with a signal aborted by {@link cancel}.
* @throws synchronously when turn-driving or another maintenance task already owns the agent.
* @returns the task promise.
*/
runMaintenance<T>(task: (signal: AbortSignal) => Promise<T>): Promise<T>
/**
* Route identified input to an inbox boundary and optionally wake the driver.
* Waking input submitted after active cancellation is queued for the next turn.
+2 -2
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 docs/core-data-structures/session.md
session.md: 0f388f5d846980b40f85815793f26fa1c32d7016
session.zh.md: 716ea383acce7aaddd4bcc604d868c07cdc38266
session.md: bbc2f3dba43a1e52b2fae39166b0cab31c55cd64
session.zh.md: 88031acf40e8fac67cbae23f4f604a8c25fe8e9a
+16 -30
View File
@@ -32,12 +32,13 @@ interface SessionEventMap {
*/
'turn/start': { turn: number }
/**
* Closes turn `turn` with the {@link TurnEndReason} that ended it. The loop
* awaits `session/flush` after an ordinary turn ends before claiming the next
* queued item. Success commits the turn; rejection is reported live and does
* not prevent later work.
* Closes turn `turn` after `step`, the last entered step (`0` when none),
* with the {@link TurnEndReason} that ended it. The loop awaits
* `session/flush` after an ordinary turn ends before claiming the next queued
* item. Success commits the turn; rejection is reported live and does not
* prevent later work.
*/
'turn/end': { turn: number; reason: TurnEndReason }
'turn/end': { turn: number; step: number; reason: TurnEndReason }
/** Opens step `step` of turn `turn` — one model call plus the tool executions it requested. */
'step/start': { turn: number; step: number }
/** Closes step `step` of turn `turn`. */
@@ -93,21 +94,14 @@ interface SessionEventMap {
*/
'request/header': { header: EpochHeader; reason: RequestHeaderReason }
/**
* Registration-bound context metadata for the route a request resolved to,
* appended inside its step beside `request/header` and only when the route
* or capacity differs from the last record. It is log-only and deliberately
* NOT part of {@link EpochHeader}: capacity is adapter metadata about a
* route, not an input the request was built from, so it must not participate
* in request reconstruction or header equality. `contextWindow` is absent
* when the route's adapter advertises no capacity.
* Route metadata for the next request, logged only when the route or capacity
* changes. It does not participate in request reconstruction or header equality.
*/
'request/context': RequestContext
/**
* Marks the end of a constructor seed. Events before it have smaller seq
* values and came from the seed (resume, fork, or replay); this lifecycle
* produced none of them. An explicitly supplied empty seed puts the marker
* at seq 0, distinguishing an empty resumed session from a fresh session.
* This log-only event is the durable projection of
* produced none of them. This log-only event is the durable projection of
* {@link Session.firstLiveSeq}. Its payload is empty — position and `time`
* carry the meaning.
*
@@ -184,17 +178,13 @@ Canonical form represents an empty system prompt or tool list as an absent field
The context metadata of the route a request resolved to is separate logged state, appended beside `request/header` inside the same step and only when the provider, model, or capacity differs from the previous record. It stays outside `EpochHeader` because that type is the reconstruction contract compared field-wise by `headerEquals`: capacity describes a route, not a request input, so folding it in would let a capacity change register as a request-envelope `change` and would pull adapter metadata into the loop's reconstruction invariant. Like `request/header`, it is not a `SurfaceEventType` and produces no LLM message. `session.requestContext()` folds the latest record incrementally. A route whose adapter advertises no capacity is recorded with `contextWindow` absent, so the new record clears an older route's capacity.
```ts type-equiv
/**
* Registration-bound context metadata of one resolved model route. Adapter
* metadata about a route rather than a request input, which is why it lives
* outside {@link EpochHeader}.
*/
/** Registration-bound metadata for one resolved model route. */
interface RequestContext {
/** Registered provider route the metadata was resolved through. */
/** Registered provider route the metadata belongs to. */
provider: string
/** Provider-owned model id the metadata belongs to. */
model: string
/** Maximum combined request and response context in tokens; absent when the adapter advertises none. */
/** Maximum combined request and response context in tokens, when advertised. */
contextWindow?: number
}
```
@@ -385,9 +375,7 @@ declare class Session {
* start here. Distinct from `header.seedLength`, the DURABLE fork-lineage
* boundary: a resumed session's constructor seed is its full stored log,
* while its header keeps the original fork value — this field is the
* in-process construction fact. An explicitly supplied empty seed has the
* same value as no seed (0); its `session/end-seed` event preserves the
* lifecycle distinction.
* in-process construction fact.
*
* Not persisted itself: a seeded session projects it into the log as the
* `session/end-seed` event, which is what a consumer reading STORED history
@@ -461,11 +449,9 @@ declare class Session {
*/
requestHeader(): EpochHeader | undefined;
/**
* The route metadata in force after the log's last `request/context` event —
* what the NEXT request deduplicates against — or undefined before any such
* record. Maintained incrementally like {@link requestHeader}, so a per-step
* read costs O(new events).
* @returns the folded context record, or undefined when none exists yet.
* Return the latest resolved route metadata, or `undefined` before the first
* `request/context` event. Each event is folded once.
* @returns the latest immutable route metadata.
*/
requestContext(): RequestContext | undefined;
/**
+16 -30
View File
@@ -32,12 +32,13 @@ interface SessionEventMap {
*/
'turn/start': { turn: number }
/**
* Closes turn `turn` with the {@link TurnEndReason} that ended it. The loop
* awaits `session/flush` after an ordinary turn ends before claiming the next
* queued item. Success commits the turn; rejection is reported live and does
* not prevent later work.
* Closes turn `turn` after `step`, the last entered step (`0` when none),
* with the {@link TurnEndReason} that ended it. The loop awaits
* `session/flush` after an ordinary turn ends before claiming the next queued
* item. Success commits the turn; rejection is reported live and does not
* prevent later work.
*/
'turn/end': { turn: number; reason: TurnEndReason }
'turn/end': { turn: number; step: number; reason: TurnEndReason }
/** Opens step `step` of turn `turn` — one model call plus the tool executions it requested. */
'step/start': { turn: number; step: number }
/** Closes step `step` of turn `turn`. */
@@ -93,21 +94,14 @@ interface SessionEventMap {
*/
'request/header': { header: EpochHeader; reason: RequestHeaderReason }
/**
* Registration-bound context metadata for the route a request resolved to,
* appended inside its step beside `request/header` and only when the route
* or capacity differs from the last record. It is log-only and deliberately
* NOT part of {@link EpochHeader}: capacity is adapter metadata about a
* route, not an input the request was built from, so it must not participate
* in request reconstruction or header equality. `contextWindow` is absent
* when the route's adapter advertises no capacity.
* Route metadata for the next request, logged only when the route or capacity
* changes. It does not participate in request reconstruction or header equality.
*/
'request/context': RequestContext
/**
* Marks the end of a constructor seed. Events before it have smaller seq
* values and came from the seed (resume, fork, or replay); this lifecycle
* produced none of them. An explicitly supplied empty seed puts the marker
* at seq 0, distinguishing an empty resumed session from a fresh session.
* This log-only event is the durable projection of
* produced none of them. This log-only event is the durable projection of
* {@link Session.firstLiveSeq}. Its payload is empty — position and `time`
* carry the meaning.
*
@@ -186,17 +180,13 @@ interface EpochHeader {
请求所解析到的路由的上下文元数据是独立的已记录状态,在同一步骤内紧随 `request/header` 追加,且仅在提供方、模型或容量与上一条记录不同时追加。它保持在 `EpochHeader` 之外,因为该类型是由 `headerEquals` 逐字段比较的重建契约:容量描述的是路由,不是请求输入,把它折叠进去会让一次容量变化被登记为请求信封的 `change`,也会把适配器元数据拉进 loop 的重建不变式。与 `request/header` 一样,它不是 `SurfaceEventType`,也不产生 LLM 消息。`session.requestContext()` 以增量方式归并最新一条记录。适配器不公布容量的路由会以缺失 `contextWindow` 的形式记录,因此新记录可以清除较早路由的容量。
```ts type-equiv
/**
* Registration-bound context metadata of one resolved model route. Adapter
* metadata about a route rather than a request input, which is why it lives
* outside {@link EpochHeader}.
*/
/** Registration-bound metadata for one resolved model route. */
interface RequestContext {
/** Registered provider route the metadata was resolved through. */
/** Registered provider route the metadata belongs to. */
provider: string
/** Provider-owned model id the metadata belongs to. */
model: string
/** Maximum combined request and response context in tokens; absent when the adapter advertises none. */
/** Maximum combined request and response context in tokens, when advertised. */
contextWindow?: number
}
```
@@ -387,9 +377,7 @@ declare class Session {
* start here. Distinct from `header.seedLength`, the DURABLE fork-lineage
* boundary: a resumed session's constructor seed is its full stored log,
* while its header keeps the original fork value — this field is the
* in-process construction fact. An explicitly supplied empty seed has the
* same value as no seed (0); its `session/end-seed` event preserves the
* lifecycle distinction.
* in-process construction fact.
*
* Not persisted itself: a seeded session projects it into the log as the
* `session/end-seed` event, which is what a consumer reading STORED history
@@ -463,11 +451,9 @@ declare class Session {
*/
requestHeader(): EpochHeader | undefined;
/**
* The route metadata in force after the log's last `request/context` event —
* what the NEXT request deduplicates against — or undefined before any such
* record. Maintained incrementally like {@link requestHeader}, so a per-step
* read costs O(new events).
* @returns the folded context record, or undefined when none exists yet.
* Return the latest resolved route metadata, or `undefined` before the first
* `request/context` event. Each event is folded once.
* @returns the latest immutable route metadata.
*/
requestContext(): RequestContext | undefined;
/**
@@ -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/system-prompt.md
system-prompt.md: 5abb8f46c13045c7d37bbe12ecf6c3744ee063b5
system-prompt.zh.md: 1088b20ba4289ad5912a193eead39d069c1a6e17
system-prompt.md: 59193c1881abcadbc8a1778cde92f5a6572eee24
system-prompt.zh.md: cc9af3d030660a1d705547b03ffbe85f7414e465
+2 -6
View File
@@ -66,15 +66,11 @@ interface PromptSection {
`PromptContext` is the cache-safe counterpart to `PromptSection`. The assembly resolves and orders these contributions, while agent-loop logs their complete current snapshot after retained model history only when it changed or compaction removed it.
```ts type-equiv
/**
* One dynamic model-context contribution. Unlike a {@link PromptSection}, its
* rendered text is materialized as a durable user-role snapshot at the request
* tail, so changing runtime state preserves the stable system/history prefix.
*/
/** Dynamic model context materialized as a durable user-role snapshot. */
interface PromptContext {
/** Unique name — a duplicate registration throws (see {@link SystemPrompt.context}). */
readonly name: string
/** Contexts are joined in ascending order, independently of system-section order. */
/** Contexts are joined in ascending order. */
readonly order: number
/** Static text or a provider evaluated for each assembly. Empty text contributes nothing. */
readonly text: string | ((context: AssembleContext) => string)
@@ -66,15 +66,11 @@ interface PromptSection {
`PromptContext` 是与 `PromptSection` 对应的缓存安全结构。组装会解析这些贡献并排序;agent loop(智能体循环)仅在完整当前快照发生变化或被压缩(compaction)移除时,才会将其记录在保留的模型历史之后。
```ts type-equiv
/**
* One dynamic model-context contribution. Unlike a {@link PromptSection}, its
* rendered text is materialized as a durable user-role snapshot at the request
* tail, so changing runtime state preserves the stable system/history prefix.
*/
/** Dynamic model context materialized as a durable user-role snapshot. */
interface PromptContext {
/** Unique name — a duplicate registration throws (see {@link SystemPrompt.context}). */
readonly name: string
/** Contexts are joined in ascending order, independently of system-section order. */
/** Contexts are joined in ascending order. */
readonly order: number
/** Static text or a provider evaluated for each assembly. Empty text contributes nothing. */
readonly text: string | ((context: AssembleContext) => string)