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

refactor(agent): return request retry action

Browse Source
This commit is contained in:
_Kerman
2026-07-27 21:17:49 +08:00
parent c6073f07c2
commit ed67ad53d2
83 changed files with 277 additions and 309 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
.agents/notes/implemented/architecture/2026-06-21-bounded-llm-request-recovery.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/architecture/2026-06-21-bounded-llm-request-recovery.md
2026-06-21-bounded-llm-request-recovery.md: dec396b992a56c50c1ccd57f04ff0e8954fed07d
2026-06-21-bounded-llm-request-recovery.zh.md: c32f2b3974d2cdb4dd8cd0f8ea132e76893f6007
2026-06-21-bounded-llm-request-recovery.md: 9c9d8a02595b988535158c9aa0ec43d6f1fa0c89
2026-06-21-bounded-llm-request-recovery.zh.md: bb9e430eaf87789452fd4cc89085d7d635f54ed1

4
.agents/notes/implemented/architecture/2026-06-21-bounded-llm-request-recovery.md
View File

@@ -6,7 +6,7 @@ English | [中文](2026-06-21-bounded-llm-request-recovery.zh.md)
## Problem
`dsh-llm` can report provider failures either by throwing during adapter dispatch or iteration or by ending with `finish { kind: 'error' | 'aborted' }`. The final adapter boundary tags thrown failures so `dsh-agent-loop` can distinguish them from middleware and result-processing defects, and the loop normalizes both delivery forms into `agent/request-error` after closing the failed step. An unhandled failure is terminal; a handling listener repairs policy-owned state, calls `agent.retry()`, and stops waterfall delegation.
`dsh-llm` can report provider failures either by throwing during adapter dispatch or iteration or by ending with `finish { kind: 'error' | 'aborted' }`. The final adapter boundary tags thrown failures so `dsh-agent-loop` can distinguish them from middleware and result-processing defects, and the loop normalizes both delivery forms into `agent/request-error` after closing the failed step. An unhandled failure is terminal; a handling listener repairs policy-owned state, returns `{ kind: 'retry' }`, and stops waterfall delegation. The [retry-action decision](../simplification/2026-07-27-request-error-retry-action.md) owns this return contract.
That boundary is already safe for another request attempt. Raw `assistant/chunk` events carry the failed `turn` and `step`, message derivation ignores them unless a successful `assistant/message` cites them, tool calls are dispatched only after a successful terminal finish and assembly, and a retry opens a new numbered turn from the durable log. The harness therefore does not need a second response lifecycle or tentative-output protocol to keep two attempts separate.
@@ -72,7 +72,7 @@ The plugin owns a lifetime `AbortController` and tracks every active backoff cal
Before sleeping, `dsh-llm-retry` appends one non-surface `llm/retry` session event containing the turn, failed step, one-based transient retry number, configured maximum, scheduled delay, and `LlmFailure`. The plugin owns the `SessionEventMap` augmentation; `dsh-session` remains generic persistence and does not absorb the optional policy's vocabulary. The event says what was scheduled, not that the next request completed; cancellation during the delay is subsequently visible on `turn/end`. The event ships only with a production renderer and replay/snapshot coverage, because its purpose is operational state rather than trace collection.
The listener calls `next()` for a non-transient code, an exhausted policy budget, or an over-cap provider delay. This preserves composition with context-overflow recovery and later policy plugins. For an owned failure it records and awaits the delay, then calls `agent.retry()` without delegating. Turn cancellation and plugin disposal end the wait without requesting a retry; the loop's cancellation/disposal checks remain authoritative.
The listener calls `next()` for a non-transient code, an exhausted policy budget, or an over-cap provider delay. This preserves composition with context-overflow recovery and later policy plugins. For an owned failure it records and awaits the delay, then returns `{ kind: 'retry' }` without delegating. Turn cancellation and plugin disposal end the wait without returning a retry; the loop's cancellation/disposal checks remain authoritative.
The agent-spine demo bundle loads the plugin so the shared stdio/TUI, one-shot CLI, and ACP example compositions use the same bounded policy. Library consumers retain explicit plugin composition: omitting the plugin leaves request failures terminal.

4
.agents/notes/implemented/architecture/2026-06-21-bounded-llm-request-recovery.zh.md
View File

@@ -6,7 +6,7 @@ Status: implemented
## 问题
`dsh-llm` 可能在适配器分发或迭代时抛出异常，也可能以 `finish { kind: 'error' | 'aborted' }` 结束，以这两种形式报告提供方失败。最终适配器边界会标记抛出的失败，使 `dsh-agent-loop` 能将其与中间件和结果处理缺陷区分开。循环关闭失败步骤后，会把两种交付形式统一规范化为 `agent/request-error`。未被处理的失败是终态；处理失败的监听器修复策略自有状态，调用 `agent.retry()`，并停止 waterfall 委托。
`dsh-llm` 可能在适配器分发或迭代时抛出异常，也可能以 `finish { kind: 'error' | 'aborted' }` 结束，以这两种形式报告提供方失败。最终适配器边界会标记抛出的失败，使 `dsh-agent-loop` 能将其与中间件和结果处理缺陷区分开。循环关闭失败步骤后，会把两种交付形式统一规范化为 `agent/request-error`。未被处理的失败是终态；处理失败的监听器修复策略自有状态，返回 `{ kind: 'retry' }`，并停止 waterfall 委托。[重试动作决策](../simplification/2026-07-27-request-error-retry-action.md)规定这一返回契约。
该边界已能安全地再次发起请求。原始 `assistant/chunk` 事件携带失败的 `turn` 和 `step`；除非某条成功的 `assistant/message` 引用这些事件，否则消息派生会忽略它们。只有终止性 finish 成功且组装完成后，系统才会分发工具调用；重试则会从持久日志开启新的编号轮次。因此，harness 无需引入第二套响应生命周期或暂定输出协议，即可分隔两次尝试。
@@ -72,7 +72,7 @@ interface Config {
休眠前，`dsh-llm-retry` 会追加一条不进入表层的 `llm/retry` 会话事件，其中包含轮次、失败步骤、从 1 开始的暂时性重试编号、已配置上限、计划延迟和 `LlmFailure`。该插件拥有 `SessionEventMap` 声明合并；`dsh-session` 继续负责通用持久化，不会吸收可选策略的词汇。事件记录已安排的内容，而不是下一个请求已完成；延迟期间取消随后会在 `turn/end` 中可见。因为该事件的目的是表示运行状态，而不是收集跟踪数据，所以它仅与生产渲染器及回放／快照覆盖一起交付。
对非暂时性 code、耗尽的策略预算或超出上限的提供方延迟，监听器会调用 `next()`。这保留了与上下文溢出恢复及后续策略插件的组合能力。对自身处理的失败，它会记录并等待延迟，然后在不委托的情况下调用 `agent.retry()`。轮次取消和插件释放会结束等待且不请求重试，此后仍以循环的取消／释放检查为准。
对非暂时性 code、耗尽的策略预算或超出上限的提供方延迟，监听器会调用 `next()`。这保留了与上下文溢出恢复及后续策略插件的组合能力。对自身处理的失败，它会记录并等待延迟，然后在不委托的情况下返回 `{ kind: 'retry' }`。轮次取消和插件释放会结束等待且不返回重试动作，此后仍以循环的取消／释放检查为准。
agent-spine 演示组合包加载该插件，因此共享的 stdio/TUI、一次性 CLI（命令行界面）和 ACP（Agent Client Protocol）示例组合使用同一有界策略。库消费方仍需显式组合插件：省略该插件时，请求失败保持终态。

4
.agents/notes/implemented/architecture/2026-07-10-after-call-compaction-pressure-and-overflow-recovery.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/architecture/2026-07-10-after-call-compaction-pressure-and-overflow-recovery.md
2026-07-10-after-call-compaction-pressure-and-overflow-recovery.md: a83408128f164f76357fa2bdf33188caaa30d8a4
2026-07-10-after-call-compaction-pressure-and-overflow-recovery.zh.md: 86037016303aa05e9f370aba3c50789665ebef1d
2026-07-10-after-call-compaction-pressure-and-overflow-recovery.md: 51d488db28c57426c75c9ed1cfc90892261c0224
2026-07-10-after-call-compaction-pressure-and-overflow-recovery.zh.md: ae33cf5c2e944e584cd3d3c6ff76d93619adf7dc

4
.agents/notes/implemented/architecture/2026-07-10-after-call-compaction-pressure-and-overflow-recovery.md
View File

@@ -24,7 +24,7 @@ The loop fires awaited serial `agent/post-step(agent, turn, step, signal)` after
`RequestError` and the `agent/request-error` waterfall represent failures after the final adapter has been selected. Each returned stream handle owns a private failure set that preserves the original thrown error identity across dispatch, iterator construction, and iteration without leaking nested-call provenance into an outer call. Terminal in-band `error` or `aborted` finishes enter the same path. Prompt assembly, request middleware, request logging, result processing, tools, step listeners, and cleanup remain ordinary failures.
The failed step closes before recovery runs. A handling listener repairs durable state, calls `agent.retry()`, and stops waterfall delegation. The loop then closes the failed turn and opens one retry turn from the durable log without an intervening idle notification. Retry policy and attempt counts remain plugin-owned; compact-basic clears its per-agent overflow count when the chain reaches terminal `agent/settled`. Both DeepSeek adapters normalize recognized provider context-limit failures to `CONTEXT_WINDOW_EXCEEDED`.
The failed step closes before recovery runs. A handling listener repairs durable state, returns `{ kind: 'retry' }`, and stops waterfall delegation. The loop then closes the failed turn and opens one retry turn from the durable log without an intervening idle notification. Retry policy and attempt counts remain plugin-owned; compact-basic clears its per-agent overflow count when the chain reaches terminal `agent/settled`. Both DeepSeek adapters normalize recognized provider context-limit failures to `CONTEXT_WINDOW_EXCEEDED`. The [retry-action decision](../simplification/2026-07-27-request-error-retry-action.md) owns the return boundary.
If cancellation lands after assistant tool calls are durable but before all calls dispatch, the loop records a synthetic `tool/call` and aborted `tool/result` pair for every undispatched call before following the normal abort path. The surface therefore never retains orphaned durable tool calls merely because cancellation won the race.
@@ -34,7 +34,7 @@ If cancellation lands after assistant tool calls are durable but before all call
For `pressure`, compact-basic resolves the durable provider/model target's adapter-owned capacity and exact-target policy, then applies the resulting threshold and retained-tail budgets to one unified `ctx.tokenMeter.measure()` result. Below pressure it returns without pruning. Once pressure qualifies, optional `ctx.toolResultPrune` rewrites oversized current results and compact-basic remeasures through the same meter; safe pressure skips the model call, while remaining pressure selects and summarizes from the pruned surface. The same singleton meter owns range pricing, provenance, shadowed token counts, and non-shrinking-summary rejection. Common defaults remain threshold ratio `0.8`, retained-history ratio `0.16`, summarization provider/model `''`, `maxTokens: 8192`, `compactionRetries: 1`, and `auto: true`; optional `modelPolicies` entries override them for an exact provider/model pair.
For canonical overflow, compact-basic requires no capacity metadata and bypasses scalar pressure and the normal retained-token budget. It prunes first, then chooses the maximal tool-balanced head range while leaving the newest indivisible unit and attempts one shrinking summary compaction under the same signal when a range exists. The automatic listener snapshots `session.surface.replaceGeneration` and calls `agent.retry()` whenever pruning or summarization increases it. This remains true when pruning lands before later summary work throws; cancellation still wins. A backend returning a result without replacement cannot authorize retry, while pruning-only progress can authorize a retry without a `CompactionResult`.
For canonical overflow, compact-basic requires no capacity metadata and bypasses scalar pressure and the normal retained-token budget. It prunes first, then chooses the maximal tool-balanced head range while leaving the newest indivisible unit and attempts one shrinking summary compaction under the same signal when a range exists. The automatic listener snapshots `session.surface.replaceGeneration` and returns `{ kind: 'retry' }` whenever pruning or summarization increases it. This remains true when pruning lands before later summary work throws; cancellation still wins. A backend returning a result without replacement cannot authorize retry, while pruning-only progress can authorize a retry without a `CompactionResult`.
`maxOverflowRetries` is optional and defaults to `1`; `0` disables overflow recovery without disabling pressure. `auto: false` registers neither automatic listener. Noncanonical errors, exhausted attempts, an already-aborted signal, a missing routed model, no safe range, no generation change, and recovery throws before any replacement all delegate to the next listener. With no later recovery, the loop reports the original provider error object and code. A recovery throw after generation advances authorizes retry from durable progress; cancellation or disposal remains authoritative even if recovery work completes concurrently.

4
.agents/notes/implemented/architecture/2026-07-10-after-call-compaction-pressure-and-overflow-recovery.zh.md
View File

@@ -24,7 +24,7 @@ Status: implemented
`RequestError` 与 `agent/request-error` waterfall 表示最终适配器已经选定之后的失败。每个返回的流句柄都绑定一个私有失败集合；该集合在分发、异步迭代器构造与迭代过程中保留原始抛出错误的身份，同时防止把嵌套调用的错误来源误归到外层调用。终止性的带内 `error` 或 `aborted` finish 进入同一路径。提示词装配、请求中间件、请求日志、结果处理、工具、step 监听器与清理仍属于普通失败。
恢复运行前，失败 step 已经关闭。负责处理的监听器修复持久状态、调用 `agent.retry()`，并停止 waterfall 委托。循环随后关闭失败 turn，并从持久日志开启一个重试 turn，中间不发布空闲通知。重试策略与尝试计数由插件自己拥有；compact-basic 在链路到达终态 `agent/settled` 时清除对应 agent 的溢出计数。两个 DeepSeek 适配器都把识别出的提供方上下文限制错误规范化为 `CONTEXT_WINDOW_EXCEEDED`。
恢复运行前，失败 step 已经关闭。负责处理的监听器修复持久状态、返回 `{ kind: 'retry' }`，并停止 waterfall 委托。循环随后关闭失败 turn，并从持久日志开启一个重试 turn，中间不发布空闲通知。重试策略与尝试计数由插件自己拥有；compact-basic 在链路到达终态 `agent/settled` 时清除对应 agent 的溢出计数。两个 DeepSeek 适配器都把识别出的提供方上下文限制错误规范化为 `CONTEXT_WINDOW_EXCEEDED`。[重试动作决策](../simplification/2026-07-27-request-error-retry-action.md)规定这一返回边界。
如果取消发生在 assistant 工具调用已经持久化之后、所有调用完成分发之前，循环会为每个尚未分发的调用记录一对合成的 `tool/call` 与 aborted `tool/result`，随后进入正常中止路径。因此，表层不会仅因取消赢得竞态而留下孤立的持久工具调用。
@@ -34,7 +34,7 @@ Status: implemented
对于 `pressure`，compact-basic 先解析持久提供方/模型目标的适配器所属容量与精确目标策略，再把得到的阈值与保留尾部预算应用到一次统一的 `ctx.tokenMeter.measure()` 结果。低于压力时直接返回，不执行剪枝。压力达到条件后，可选的 `ctx.toolResultPrune` 会改写当前表层中过大的工具结果，compact-basic 再通过同一个 meter 重新计量；若压力恢复安全则跳过模型调用，否则从已剪枝表层选择范围并生成摘要。范围定价、来源、被遮蔽 token 数与非缩小摘要拒绝也由同一个单例 meter 完成。通用默认值保持为阈值比例 `0.8`、保留历史比例 `0.16`、摘要提供方/模型 `''`、`maxTokens: 8192`、`compactionRetries: 1` 与 `auto: true`；可选 `modelPolicies` 项可以按精确提供方/模型组合覆盖这些值。
对于规范化溢出，compact-basic 不要求容量元数据，并绕过标量压力与普通保留 token 预算。它先执行剪枝，再在保留最新不可分割单元的同时选择最大的工具配对平衡头部范围；存在范围时，才在同一 signal 下尝试一次缩小摘要压缩。自动监听器先记录 `session.surface.replaceGeneration`，剪枝或摘要让 generation 增加时就调用 `agent.retry()`。即使剪枝先落盘而后续摘要工作抛错，这条规则仍然成立；取消依然优先。后端若只返回结果但没有替换表层，不能授权重试；只有剪枝取得进展时，即使没有 `CompactionResult` 也可以授权重试。
对于规范化溢出，compact-basic 不要求容量元数据，并绕过标量压力与普通保留 token 预算。它先执行剪枝，再在保留最新不可分割单元的同时选择最大的工具配对平衡头部范围；存在范围时，才在同一 signal 下尝试一次缩小摘要压缩。自动监听器先记录 `session.surface.replaceGeneration`，剪枝或摘要让 generation 增加时就返回 `{ kind: 'retry' }`。即使剪枝先落盘而后续摘要工作抛错，这条规则仍然成立；取消依然优先。后端若只返回结果但没有替换表层，不能授权重试；只有剪枝取得进展时，即使没有 `CompactionResult` 也可以授权重试。
`maxOverflowRetries` 可选且默认为 `1`；`0` 只禁用溢出恢复，不会禁用压力检查。`auto: false` 不注册任何自动监听器。非规范化错误、尝试耗尽、已经中止的 signal、缺失路由模型、没有安全范围、generation 未变化，以及在任何替换之前恢复抛错，都会委托给下一个监听器。若没有后续恢复，循环报告原始提供方错误对象与代码。generation 增加后的恢复抛错会基于持久进展授权重试；即使恢复工作并发完成，取消或销毁仍具有最终优先级。

4
.agents/notes/implemented/architecture/2026-07-24-separate-context-injection-from-turn-execution.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/architecture/2026-07-24-separate-context-injection-from-turn-execution.md
2026-07-24-separate-context-injection-from-turn-execution.md: 37798795b706719ca46a4495ef3f82a7968c501f
2026-07-24-separate-context-injection-from-turn-execution.zh.md: 560f97d00c491adad4deeeeee8c8ed7ea8b88028
2026-07-24-separate-context-injection-from-turn-execution.md: b74cd6bdc48e795e57d780ab31a907ffe94dd518
2026-07-24-separate-context-injection-from-turn-execution.zh.md: f2421d2fc7b8c1329dd1349a6fb088407ac5fc75

2
.agents/notes/implemented/architecture/2026-07-24-separate-context-injection-from-turn-execution.md
View File

@@ -30,7 +30,7 @@ During prompt admission or an open turn, `inject()` stages context in the loop o
Outside that window, `inject()` appends its `user/message` immediately. It does not increment turn numbering, emit `turn/start` or `turn/end`, change agent status, or run the model; persistence observes the append through `session/event`.
If prompt admission blocks or fails, a caller-staged context-only batch appends immediately without a turn. Steering and context staged beside it remain in the outbox for `retry()` or a later admitted prompt; cancellation or disposal may discard them. Hook-produced `additionalContexts` never materialize because they belong to the rejected admission decision.
If prompt admission blocks or fails, a caller-staged context-only batch appends immediately without a turn. Steering and context staged beside it remain in the outbox for a later admitted prompt; cancellation or disposal may discard them. Hook-produced `additionalContexts` never materialize because they belong to the rejected admission decision.
The session invariant permits `user/message` between turns while continuing to require turn enclosure for execution events, steering, assistant output, tools, and package-added events by default. Persistence, recovery, resume, fork, and compaction treat a valid out-of-turn `user/message` as committed session history rather than an interrupted or discardable turn tail.

2
.agents/notes/implemented/architecture/2026-07-24-separate-context-injection-from-turn-execution.zh.md
View File

@@ -30,7 +30,7 @@ agent API 曾用三种相互重叠的方式表示面向模型的补充输入：
在该窗口之外，`inject()` 会立即追加对应的 `user/message`。它不会增加轮次编号、发出 `turn/start` 或 `turn/end`、改变 agent 状态，也不会运行模型；持久化通过 `session/event` 观察这次追加。
如果提示词准入被阻止或失败，调用方暂存的仅含上下文的批次会立即追加，且不产生轮次。steering 及与其一同暂存的上下文会留在 outbox 中，供 `retry()` 或后续获准提示词使用；取消或 dispose（资源释放）可能丢弃它们。钩子产生的 `additionalContexts` 属于被拒绝的准入决策，因此永远不会落入日志。
如果提示词准入被阻止或失败，调用方暂存的仅含上下文的批次会立即追加，且不产生轮次。steering 及与其一同暂存的上下文会留在 outbox 中，供后续获准提示词使用；取消或 dispose（资源释放）可能丢弃它们。钩子产生的 `additionalContexts` 属于被拒绝的准入决策，因此永远不会落入日志。
会话不变量允许 `user/message` 位于两个轮次之间，同时继续要求执行事件、steering、助手输出、工具事件以及默认的包扩展事件均受轮次边界约束。持久化、恢复、resume、fork 和压缩会把合法的轮次外 `user/message` 当作已提交会话历史，而不是中断轮次或可丢弃的日志尾部。

6
.agents/notes/implemented/feature/2026-06-18-compaction-capability-seam.i18n.yaml
View File

@@ -1,6 +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
2026-06-18-compaction-capability-seam.md: 73fd78af9d4ef86ae7bafc3a3608ff1afbbd87bd
2026-06-18-compaction-capability-seam.zh.md: 8fc92ccd5ae6ad61a174e01b3ee3afa1df0a4529
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-06-18-compaction-capability-seam.md
2026-06-18-compaction-capability-seam.md: 3c219b734e148b963fb5857de89c16f28c2bd402
2026-06-18-compaction-capability-seam.zh.md: b2c7e9720b596705b60a284e6ccf1448a782b7fc

2
.agents/notes/implemented/feature/2026-06-18-compaction-capability-seam.md
View File

@@ -39,7 +39,7 @@ An earlier draft put the full algorithm (the retention walk, token-summing, text
Successful-call pressure cannot run at pre-step because final `agent/request` routing, provider output, tool results, buffered context, and steering do not exist there. Serial `agent/post-step(agent, turn, step, signal)` fires after those facts are durable and before `step/end`. `dsh-compact-basic` measures the canonical logged request through `ctx.tokenMeter`, so the next request sees any replacement without a speculative envelope override. Once pressure qualifies, optional `ctx.toolResultPrune` rewriting runs before summary selection; compact-basic remeasures the durable surface and skips summarization if pruning restores safe pressure.
Canonical provider context overflow takes a separate path. The failed step closes and `agent/request-error` receives the original request error. Compact-basic owns its per-agent overflow count, prunes before forcing one useful balanced reduction, and calls `agent.retry()` only if `session.surface.replaceGeneration` increases, including pruning-only progress when no summary range exists. The loop then closes the failed turn, opens a new numbered retry turn, and reconstructs its request from the durable log. No replacement, a recovery failure before any replacement, cancellation, an exhausted cap, or an unrelated error preserves the original provider failure. If pruning already advanced the generation before later summary work fails, recovery retries from that durable pruned surface unless cancellation or disposal wins. The complete lifecycle decision is in the [after-call recovery Agent Note](../architecture/2026-07-10-after-call-compaction-pressure-and-overflow-recovery.md).
Canonical provider context overflow takes a separate path. The failed step closes and `agent/request-error` receives the original request error. Compact-basic owns its per-agent overflow count, prunes before forcing one useful balanced reduction, and returns `{ kind: 'retry' }` only if `session.surface.replaceGeneration` increases, including pruning-only progress when no summary range exists. The loop then closes the failed turn, opens a new numbered retry turn, and reconstructs its request from the durable log. No replacement, a recovery failure before any replacement, cancellation, an exhausted cap, or an unrelated error preserves the original provider failure. If pruning already advanced the generation before later summary work fails, recovery retries from that durable pruned surface unless cancellation or disposal wins. The complete lifecycle decision is in the [after-call recovery Agent Note](../architecture/2026-07-10-after-call-compaction-pressure-and-overflow-recovery.md).
```
assistant/message → tool/result/context/steering

2
.agents/notes/implemented/feature/2026-06-18-compaction-capability-seam.zh.md
View File

@@ -39,7 +39,7 @@ Status: implemented
成功调用的压力检查不能在步骤前运行，因为最终的 `agent/request` 路由、提供方输出、工具结果、缓冲上下文与 steering 当时尚不存在。串行的 `agent/post-step(agent, turn, step, signal)` 会在这些事实持久化后、`step/end` 之前触发。`dsh-compact-basic` 通过 `ctx.tokenMeter` 测量规范的已记录请求，因此下一个请求无需推测性覆盖信封即可看到任何替换。压力达到条件后，可选的 `ctx.toolResultPrune` 重写在摘要范围选择前运行；compact-basic 重新测量持久 surface，如果修剪恢复到安全压力便跳过摘要生成。
规范的提供方上下文溢出走另一条路径。失败步骤先关闭，`agent/request-error` 接收原始请求错误。compact-basic 自行持有按 agent 计的溢出次数，在强制执行一次有效且平衡的缩减前先修剪，且仅当 `session.surface.replaceGeneration` 增加时才调用 `agent.retry()`；这包括没有摘要范围时仅修剪取得的进展。随后循环关闭失败轮次，开启新的编号重试轮次，并从持久日志重建请求。没有替换、任何替换前的恢复失败、取消、耗尽的上限或无关错误都会保留原始提供方失败。如果修剪已经推进 generation，而后续摘要工作失败，恢复会从该持久的已修剪 surface 重试，除非取消或资源释放胜出。完整生命周期决策见[调用后恢复 Agent Note](../architecture/2026-07-10-after-call-compaction-pressure-and-overflow-recovery.md)。
规范的提供方上下文溢出走另一条路径。失败步骤先关闭，`agent/request-error` 接收原始请求错误。compact-basic 自行持有按 agent 计的溢出次数，在强制执行一次有效且平衡的缩减前先修剪，且仅当 `session.surface.replaceGeneration` 增加时才返回 `{ kind: 'retry' }`；这包括没有摘要范围时仅修剪取得的进展。随后循环关闭失败轮次，开启新的编号重试轮次，并从持久日志重建请求。没有替换、任何替换前的恢复失败、取消、耗尽的上限或无关错误都会保留原始提供方失败。如果修剪已经推进 generation，而后续摘要工作失败，恢复会从该持久的已修剪 surface 重试，除非取消或资源释放胜出。完整生命周期决策见[调用后恢复 Agent Note](../architecture/2026-07-10-after-call-compaction-pressure-and-overflow-recovery.md)。
```
assistant/message → tool/result/context/steering

6
.agents/notes/implemented/simplification/2026-07-17-one-send-one-turn.i18n.yaml
View File

@@ -1,6 +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
2026-07-17-one-send-one-turn.md: d6c771a2b87e14c52b05c9cd8accaa44405fba16
2026-07-17-one-send-one-turn.zh.md: a057ebce38208f0ed6be3b8b97dd4c434dd3e721
# pnpm run verify-translation-pairing --write .agents/notes/implemented/simplification/2026-07-17-one-send-one-turn.md
2026-07-17-one-send-one-turn.md: dcc6c0aa483a0e53205dbaeef4e2b903f5f6a215
2026-07-17-one-send-one-turn.zh.md: 8c12481defe6608c13ee81132b432b4d8b17b681

2
.agents/notes/implemented/simplification/2026-07-17-one-send-one-turn.md
View File

@@ -22,7 +22,7 @@ If messages A and B are both processed, B's turn starts only after A records `tu
Prompt admission decides one message at a time before a turn opens. An allowed prompt becomes that turn's `user/message`; a blocked prompt is discarded without opening a turn or writing session history. Mixed-batch and all-blocked-batch branches do not exist.
The no-batching rule applies only to ordinary `send()`. Running `steer()` puts input in the outbox. While a turn remains open, the loop records that input at the next step boundary and steering makes another step the default. A failure before that boundary leaves the steering staged without waking the agent; `retry()` or a later prompt takes it, while cancellation or disposal can discard it. When the agent is idle, `steer()` delegates to `send()`, so it creates an independent ordinary queue item.
The no-batching rule applies only to ordinary `send()`. Running `steer()` puts input in the outbox. While a turn remains open, the loop records that input at the next step boundary and steering makes another step the default. A failure before that boundary leaves the steering staged without waking the agent; a request-error retry action or a later prompt takes it, while cancellation or disposal can discard it. When the agent is idle, `steer()` delegates to `send()`, so it creates an independent ordinary queue item.
`inject()` continues to add model-facing context without submitting an ordinary message. During a turn it waits in the outbox for a safe step boundary; while idle it appends a `user/message` directly, without opening a turn or running the model. Persistence owns the resulting eager drain. `cancel()` remains a whole-agent operation that can clear all unstarted ordinary and steering input and abort the current step. `status` and `whenIdle()` also describe the whole agent, not one message. Several one-message turns can share one `running` interval, so `running` does not prove that a turn is open.

2
.agents/notes/implemented/simplification/2026-07-17-one-send-one-turn.zh.md
View File

@@ -22,7 +22,7 @@ Status: implemented
提示词准入会在轮次打开前，每次只决定一条消息。获准提示词成为该轮次的 `user/message`；被阻止的提示词会被丢弃，不打开轮次，也不写入会话历史。实现中不存在混合批次或全阻止批次分支。
上述不合批规则只适用于普通 `send()`。agent 运行时，`steer()` 会把输入放入 outbox。只要当前轮次仍然打开，agent loop 就会在下一个步骤边界记录该输入，而 steering（中途引导）会默认让循环再执行一个步骤。在到达该边界前发生失败，会让 steering 保持暂存且不唤醒 agent；`retry()` 或后续提示词会取走它，而取消或 dispose 可以将其丢弃。agent 空闲时，`steer()` 委托给 `send()`，因此会创建一个独立的普通队列项。
上述不合批规则只适用于普通 `send()`。agent 运行时，`steer()` 会把输入放入 outbox。只要当前轮次仍然打开，agent loop 就会在下一个步骤边界记录该输入，而 steering（中途引导）会默认让循环再执行一个步骤。在到达该边界前发生失败，会让 steering 保持暂存且不唤醒 agent；请求错误的重试动作或后续提示词会取走它，而取消或 dispose 可以将其丢弃。agent 空闲时，`steer()` 委托给 `send()`，因此会创建一个独立的普通队列项。
`inject()` 继续添加面向模型的上下文，而不提交普通消息。轮次打开时，该上下文会留在 outbox 中，等待安全的步骤边界；agent 空闲时，系统会直接追加一条 `user/message`，既不打开轮次，也不运行模型。持久化层独立负责由此产生的即时排空。`cancel()` 仍是面向整个 agent 的操作，可以清空所有尚未启动的普通输入和 steering，并中止当前步骤。`status` 和 `whenIdle()` 描述的也是整个 agent，而不是某一条消息。多个单消息轮次可以共用一个 `running` 区间，因此 `running` 不表示轮次一定处于打开状态。

4
.agents/notes/implemented/simplification/2026-07-24-agent-loop-observable-state-machine.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/simplification/2026-07-24-agent-loop-observable-state-machine.md
2026-07-24-agent-loop-observable-state-machine.md: 2905e003a05bf1c8caa8f784787fb76f8e76b7d7
2026-07-24-agent-loop-observable-state-machine.zh.md: 1d6b5e5d34d020ef5561b8016b5ad065c96bff12
2026-07-24-agent-loop-observable-state-machine.md: 54730de8aa73342b609d423dc478edb076d7844b
2026-07-24-agent-loop-observable-state-machine.zh.md: 206ac701472f14823300df0c812f2cc818f852f5

8
.agents/notes/implemented/simplification/2026-07-24-agent-loop-observable-state-machine.md
View File

@@ -25,9 +25,9 @@ The loop keeps five machine extension events. `agent/prompt-submit` admits, rewr
Continuation and termination are data rather than returned control enums. Tool calls and accepted steering require another step. A tool result carrying `concludesTurn` ends the tool loop at its step. The loop does not expose general `ContinuationDecision` or terminal-stop return channels.
A model-request failure closes its step, then enters `agent/request-error` with the exact error, normalized `LlmFailure`, and live turn signal. A listener that owns recovery repairs state, calls `agent.retry()`, and returns without delegating. The loop closes the failed turn and opens one retry turn over that state without an intervening idle notification; retry is not another step inside the failed turn. `agent/settled` reports the terminal outcome, and `agent/error` remains the live error notification for consumers that report failures independently of turn settlement.
A model-request failure closes its step, then enters `agent/request-error` with the exact error, normalized `LlmFailure`, and live turn signal. A listener that owns recovery repairs state, returns `{ kind: 'retry' }`, and stops delegating. The loop closes the failed turn and opens one retry turn over that state without an intervening idle notification; retry is not another step inside the failed turn. `agent/settled` reports the terminal outcome, and `agent/error` remains the live error notification for consumers that report failures independently of turn settlement. The [retry-action decision](2026-07-27-request-error-retry-action.md) supersedes the command-shaped part of this design.
The event taxonomy removes `agent/pre-step`, `agent/post-step`, `agent/session-prefix`, `agent/step-result`, `agent/turn-continuation`, and `agent/turn-stop`. Durable turn and step boundaries remain session events. Model-facing additions use logged message channels, request configuration uses `agent/request`, response content is recorded as assembled, failed-request recovery uses `agent/request-error` plus `agent.retry()`, and end-of-turn continuation uses `agent/turn-stopping` plus steering.
The event taxonomy removes `agent/pre-step`, `agent/post-step`, `agent/session-prefix`, `agent/step-result`, `agent/turn-continuation`, and `agent/turn-stop`. Durable turn and step boundaries remain session events. Model-facing additions use logged message channels, request configuration uses `agent/request`, response content is recorded as assembled, failed-request recovery uses the `agent/request-error` return action, and end-of-turn continuation uses `agent/turn-stopping` plus steering.
## Alternatives considered
@@ -35,7 +35,7 @@ The event taxonomy removes `agent/pre-step`, `agent/post-step`, `agent/session-p
**Represent disposal as a third `AgentStatus`.** This gives retained handles a terminal status value but duplicates the registry lifecycle already expressed by `agent/disposed`. The decision keeps `AgentStatus` about live activity and makes registration lifetime a separate dimension.
**Return a retry decision from `agent/request-error`.** A returned instruction duplicates the existing `agent.retry()` command and requires the loop to carry policy history across attempts. The waterfall remains useful for ordered ownership: an unhandled listener delegates, while a handling listener performs its awaited repair, calls `agent.retry()`, and stops delegation.
**Return a retry decision from `agent/request-error`.** This alternative is superseded by the [retry-action decision](2026-07-27-request-error-retry-action.md), which removes the duplicate command and keeps the decision local to the waterfall result.
**Mirror durable turn and step boundaries as agent events.** This gives live consumers a second event stream for the same facts. The decision keeps the session log as the source of truth and exposes only extension checkpoints or live-only facts that the durable stream cannot carry.
@@ -45,7 +45,7 @@ The observable machine is smaller and compositional: registration lifetime, acti
Plugins no longer rewrite every phase of the loop. There is no request-only message prefix, assistant-message transform, post-step checkpoint, generic continuation enum, generic terminal-stop result, or in-turn request retry. Extensions use the remaining owned channels instead of recreating those phases.
Continuation plugins publish durable steering rather than returning an unlogged reason. Recovery plugins act after the failed step and explicitly schedule another turn through `agent.retry()`. This makes every attempt a complete turn while keeping asynchronous repair and policy ownership at one narrow waterfall boundary.
Continuation plugins publish durable steering rather than returning an unlogged reason. Recovery plugins act after the failed step and return an explicit retry action. This makes every attempt a complete turn while keeping asynchronous repair and policy ownership at one narrow waterfall boundary.
The inbox lifecycle complements, rather than replaces, the durable session log. `AgentMessageId` correlates acceptance with claim or discard; turn and step numbers, messages, tool activity, and terminal reasons remain session facts.

8
.agents/notes/implemented/simplification/2026-07-24-agent-loop-observable-state-machine.zh.md
View File

@@ -25,9 +25,9 @@ agent 生命周期、agent 整体活动状态、收件箱条目的进度以及
是否继续和终止执行由数据表达，不再由返回的控制枚举表达。工具调用和已接受的 steering 要求再执行一个步骤。携带 `concludesTurn` 的工具结果会在其所属步骤终止工具循环。循环不再暴露通用的 `ContinuationDecision` 或终止停止返回通道。
模型请求失败会先关闭当前步骤，再携带准确错误、标准化 `LlmFailure` 和仍有效的轮次信号进入 `agent/request-error`。负责恢复的监听器修复状态、调用 `agent.retry()`，并停止继续委托。循环会关闭失败轮次，并基于该状态开启一个重试轮次，中间不发布空闲通知；重试不是失败轮次内的另一个步骤。`agent/settled` 报告终态结果；对于需要脱离轮次结算单独报告失败的消费方，`agent/error` 仍作为实时错误通知保留。
模型请求失败会先关闭当前步骤，再携带准确错误、标准化 `LlmFailure` 和仍有效的轮次信号进入 `agent/request-error`。负责恢复的监听器修复状态、返回 `{ kind: 'retry' }`，并停止继续委托。循环会关闭失败轮次，并基于该状态开启一个重试轮次，中间不发布空闲通知；重试不是失败轮次内的另一个步骤。`agent/settled` 报告终态结果；对于需要脱离轮次结算单独报告失败的消费方，`agent/error` 仍作为实时错误通知保留。[重试动作决策](2026-07-27-request-error-retry-action.md)取代了本设计中命令形式的部分。
事件分类体系移除了 `agent/pre-step`、`agent/post-step`、`agent/session-prefix`、`agent/step-result`、`agent/turn-continuation` 和 `agent/turn-stop`。持久的轮次与步骤边界仍由会话事件记录。面向模型的新增内容使用有日志记录的消息通道，请求配置使用 `agent/request`，响应内容按组装后的原样记录，失败请求恢复使用 `agent/request-error` 加 `agent.retry()`，轮次结束时是否继续则使用 `agent/turn-stopping` 加 steering 表达。
事件分类体系移除了 `agent/pre-step`、`agent/post-step`、`agent/session-prefix`、`agent/step-result`、`agent/turn-continuation` 和 `agent/turn-stop`。持久的轮次与步骤边界仍由会话事件记录。面向模型的新增内容使用有日志记录的消息通道，请求配置使用 `agent/request`，响应内容按组装后的原样记录，失败请求恢复使用 `agent/request-error` 返回动作，轮次结束时是否继续则使用 `agent/turn-stopping` 加 steering 表达。
## 考虑过的替代方案
@@ -35,7 +35,7 @@ agent 生命周期、agent 整体活动状态、收件箱条目的进度以及
**将 dispose 表示为第三种 `AgentStatus`。** 这样会让仍被持有的句柄得到一个终止状态值，但也会重复表达 `agent/disposed` 已经体现的注册表生命周期。当前决策让 `AgentStatus` 只表示活动中 agent 的状态，并将注册生命周期作为独立维度。
**让 `agent/request-error` 返回重试决策。** 返回指令会与现有的 `agent.retry()` 命令重复，还要求循环跨尝试携带策略历史。Waterfall 只需保留有序归属能力：未处理的监听器继续委托，负责处理的监听器完成可等待的修复、调用 `agent.retry()`，然后停止委托。
**让 `agent/request-error` 返回重试决策。** 这一替代方案已由[重试动作决策](2026-07-27-request-error-retry-action.md)取代；新决策移除了重复命令，并将决策局限于 waterfall 的返回结果。
**将持久的轮次与步骤边界映射为 agent 事件。** 这样会为同一事实向实时消费方提供第二条事件流。当前决策将会话日志保留为真源，仅暴露扩展检查点或持久事件流无法承载的纯实时事实。
@@ -45,7 +45,7 @@ agent 生命周期、agent 整体活动状态、收件箱条目的进度以及
插件不再能够改写循环的每个阶段。不再提供仅用于请求的消息前缀、助手消息变换、步骤后检查点、通用的继续执行枚举、通用的终止停止结果或轮次内请求重试。扩展改用剩余的归属明确的通道，而不是重新构造这些阶段。
负责继续执行的插件发布可持久化的 steering，而不是返回未记录到日志中的原因。恢复插件在失败步骤结束后处理错误，并通过 `agent.retry()` 显式安排另一个轮次。这样，每次尝试都会成为完整轮次，同时异步修复和策略归属集中在一个狭窄的 waterfall 边界。
负责继续执行的插件发布可持久化的 steering，而不是返回未记录到日志中的原因。恢复插件在失败步骤结束后处理错误，并返回显式重试动作。这样，每次尝试都会成为完整轮次，同时异步修复和策略归属集中在一个狭窄的 waterfall 边界。
收件箱生命周期用于补充持久会话日志，而非取代它。`AgentMessageId` 将接受操作与领取或丢弃操作关联起来；轮次编号与步骤编号、消息、工具活动和终止原因仍属于会话事实。

6
.agents/notes/implemented/simplification/2026-07-27-request-error-retry-action.i18n.yaml Normal file
View File

@@ -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/simplification/2026-07-27-request-error-retry-action.md
2026-07-27-request-error-retry-action.md: 3057b9fa28cf203c9374930fe97421918b4c1a6f
2026-07-27-request-error-retry-action.zh.md: bcb4e592f0c3d86f896e279cf0e3ea400741a1bb

29
.agents/notes/implemented/simplification/2026-07-27-request-error-retry-action.md Normal file
View File

@@ -0,0 +1,29 @@
# Agent Note: Request-error retry action
Status: implemented
English | [中文](2026-07-27-request-error-retry-action.zh.md)
## Problem
Model-request recovery was decided inside `agent/request-error` but communicated through `Agent.retry()`. That public command was valid during one narrow waterfall window and while idle, rejected other running states, and required `ReactLoopAgent` to retain a mutable retry window beside the waterfall result. The recovery plugins were the only production callers, so the wider live-agent capability exposed states and behavior unrelated to their policy decision.
## Decision
`agent/request-error` returns `RequestErrorAction`, whose handling action is `{ kind: 'retry' }`; the default `undefined` keeps the failed turn terminal. A listener that does not own the failure calls `next()`. A listener that owns it performs any awaited repair and returns the retry action without delegating.
The loop reads the action after the waterfall settles, closes the failed turn, and opens one retry turn from durable history. It rechecks the turn signal when consuming the action, so cancellation or disposal during recovery prevents the retry even if a listener returns it afterward. A thrown recovery never produces an action.
`Agent` and `ReactLoopAgent` expose no `retry()` method. Ordinary new work enters through `send()` and its `followup()`, `steer()`, and `inject()` presets; only a handled model-request failure can open a promptless retry turn.
## Alternatives considered
**Keep `Agent.retry()` as the recovery command.** Runtime guards can restrict the command to the request-error window, but the interface still advertises an idle resummon operation with no production consumer and the loop still needs mutable side-channel state to recover a decision already owned by the waterfall.
**Return an explicit terminal action.** `undefined` already represents the waterfall's unhandled default and composes directly through `next()`. A second `{ kind: 'fail' }` value would add no distinct behavior or ownership information.
## Consequences
Recovery ownership, asynchronous repair, and the retry decision share one typed return path. The live-agent interface and concrete loop lose the idle resummon capability and retry-window state. Callers cannot restart arbitrary failed non-request work without submitting a later prompt, while transient and context-overflow policies retain numbered retry turns, durable-history reconstruction, finite private budgets, and cancellation precedence.
Focused agent-loop tests pin retry chaining, terminal fallthrough, recovery failure, and cancellation races. The llm-retry and compact-basic suites pin their policy-owned action returns, and the ACP, goal-session, and plan-mode integrations pin successor-turn adoption.

29
.agents/notes/implemented/simplification/2026-07-27-request-error-retry-action.zh.md Normal file
View File

@@ -0,0 +1,29 @@
# Agent Note: 请求错误重试动作
Status: implemented
[English](2026-07-27-request-error-retry-action.md) | 中文
## 问题
模型请求恢复由 `agent/request-error` 内部决定，却通过 `Agent.retry()` 传达。这个公开命令只在一个狭窄的 waterfall（瀑布式事件）窗口内和空闲时有效，在其他运行状态下会被拒绝，并要求 `ReactLoopAgent` 在 waterfall 结果旁保留一个可变的重试窗口。恢复插件是仅有的生产调用方，因此更宽泛的活跃 agent（智能体）能力暴露了与其策略决策无关的状态与行为。
## 决策
`agent/request-error` 返回 `RequestErrorAction`，其中负责处理的动作是 `{ kind: 'retry' }`；默认的 `undefined` 会让失败轮次保持终态。不拥有该失败的监听器调用 `next()`。拥有该失败的监听器执行所有需要等待的修复，然后直接返回重试动作而不继续委托。
waterfall 结算后，循环读取该动作，关闭失败轮次，并从持久历史开启一个重试轮次。循环在使用该动作时会再次检查轮次信号，因此即使监听器随后返回重试动作，恢复期间发生的取消或资源释放仍会阻止重试。抛出异常的恢复不会产生动作。
`Agent` 与 `ReactLoopAgent` 均不暴露 `retry()` 方法。普通新工作通过 `send()` 及其 `followup()`、`steer()` 和 `inject()` 预设进入；只有已处理的模型请求失败才能开启没有提示词的重试轮次。
## 曾考虑的替代方案
**保留 `Agent.retry()` 作为恢复命令。** 运行时防护检查可以将该命令限制在请求错误窗口内，但接口仍会暴露一个没有生产消费方的空闲无提示词再运行操作，循环也仍需通过可变的旁路状态恢复已经由 waterfall 决定的结果。
**返回显式终态动作。** `undefined` 已经表示 waterfall 未处理时的默认值，并可直接通过 `next()` 组合。再添加一个 `{ kind: 'fail' }` 值不会提供不同的行为或归属信息。
## 后果
恢复归属、异步修复和重试决策共用一条类型化返回路径。活跃 agent 接口与具体循环不再具有空闲无提示词再运行能力和重试窗口状态。调用方如果不提交后续提示词，就无法重启任意失败的非请求工作；瞬时策略与上下文溢出策略则保留编号重试轮次、从持久历史重建、有限的策略私有预算和取消优先级。
聚焦的 agent-loop 测试固定了重试链、未处理失败保持终态、恢复失败和取消竞态。llm-retry 与 compact-basic 测试套件固定其策略自有的动作返回，而 ACP、goal-session 和 plan-mode 集成测试固定后继轮次承接。