Integrate NyxID tools and workflow observation support#1698
Open
eanzhao wants to merge 146 commits into
Open
Conversation
Replaces magic-string suspension_type on WorkflowSuspendedEvent with a proto enum and adds expected_options to the event itself, eliminating the string-compare switch in WorkflowHumanInteractionProjector. Modules emit the typed enum and seed default options via a single helper. Issue #553 finding 5 (rollout step 1). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
通过强类型 tool execution port 统一 workflow 直调工具执行路径，避免缺失审批管线时继续裸执行工具。 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
覆盖 timeout 与 pending approval 通过执行端口映射 typed status，避免 workflow direct tool_call 回归到绕过审批链路。 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…tool-call-approval 修复 workflow tool_call 审批中间件旁路
Remove the lossy runtime LLM override mirror so llm_call derives its child request from the workflow ToolContext single source of truth. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
合并 #1664：修复 workflow llm_call ToolContext 传播。 Review-gate: architect/tests/quality 均 approve。 CI: all required checks completed successfully. ⟦AI:AUTO-LOOP⟧
eanzhao PR #1691 -> feature/integrate integration.
将 eanzhao PR #554 合入 feature/integrate。 冲突解决： - workflow_execution_messages.proto: dev 的 secure(10)/redacted_output(11) 与 PR 的 suspension_type/expected_options 字段号冲突；保留 dev 已部署的 10/11，PR 新字段移至 suspension_type=12 / expected_options=13；slot 3 保持 reserved（旧字符串 suspension_type 已移除）。 - EventEnvelopeToWorkflowRunEventMapper.cs: 合并 dev 的 secure-input 处理(Secure/RedactedOutput/ resolved variableName/filtered metadata) 与 PR 的 enum→ToWireName + Options。 - WorkflowHumanInteractionProjector.cs: 保留 dev 的 BuildAnnotations，删除被 enum 取代的死方法 ResolveOptions。 - demos/Aevatar.Demos.Workflow.Web/Program.cs: dev 已删除该 demo，保持删除。 - 三个 Host.Api 测试 + 一个集成测试：WorkflowSuspendedEvent.SuspensionType 字符串字面量改为枚举值。 验证：build 0 error；Host.Api mapper/projector 47 测试通过；SecureInputModule 集成测试通过。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- ToLegacyMetadata() 与 AgentToolRequestContext.LegacyMetadata 增加边界约束注释： 仅限旧/第三方 tool adapter 边界调用；主链必须消费 typed AgentToolExecutionContext 字段。 - 重命名 CurrentMetadata -> LegacyMetadata（保留 CurrentMetadata 别名兼容旧测试）。 - AIGAgentBase: 移除从 legacy metadata 读取 MaxToolRoundsOverride 的主链回退 （删除 ResolveMaxToolRounds），改用 EffectiveConfig.MaxToolRounds + typed Routing.MaxToolRoundsOverride。 - architecture_guards.sh: 扩展 guard 禁止主链消费 LegacyMetadata/CurrentMetadata/TryGet/ToLegacyMetadata （adapter/test 除外）。 - 新增测试覆盖允许边界与 guard 存在性。 验证：dotnet build aevatar.slnx 0 error；Aevatar.AI.Tests 741 passed；guard clean。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- 新增 CQRS 交互层统一 seam ICommandObservationScopeLeasePreparation<...>，由 DefaultCommandInteractionService 在 dispatch 前统一准备 observation lease。 - Workflow 与 GAgent draft-run Application service 不再各自调用 projection EnsureAsync/ActivateAsync； feature 适配器收窄并改名为 *ObservationScopeLeasePreparationPort。 - 澄清 ProjectionUnavailable ACK 语义：表示「本次交互无法建立 observation」，与 target-missing、 dispatch 失败明确区分。 - 测试：新增 CQRS-core prepare-fail / prepare-release / 顺序覆盖；更新 Workflow/GAgent 测试； 删除随实现移除的 per-service activation 旧测试。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…tion.cs .gitignore 的 `runs/` 规则在 macOS 大小写不敏感文件系统上误匹配 src/workflow/Aevatar.Workflow.Application/Runs/，导致该新文件被 git add -A 跳过， 合并后 trunk 编译缺类。force-add 补回（已 tracked 文件不受 gitignore 影响；Linux CI 不受此规则误伤）。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- 新增 typed views（ToolResultView / SkillSearchToolResultView / SkillLoadToolResultView / SkillSearchMatchView，位于 AI.Abstractions.LLMProviders；进程内 intra-turn 控制流视图，未跨 actor/session 序列化边界，故不引入 Protobuf）。 - OrnnSearchSkillsTool / UseSkillTool 输出结构化 JSON envelope（result_type + status + matches/error + human-readable text）。 - ToolCallLoop.BuildToolResultMessage 在边界一次性解析 envelope -> typed view 挂到 ChatMessage， Content 取 envelope 的 text（LLM 仍看可读文本，非 raw JSON）；非 skill 工具不受影响；旧字符串输出仅在边界兜底。 - SkillRecoveryPlanner 仅依据 typed 字段判定 search 命中/无匹配/错误、skill name、load 成功/失败/not-found。 - 测试：重写 SkillRecoveryPlannerTests 等，覆盖无匹配/命中/错误/加载成功失败/文本变化但 typed result 不变。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…odel 轮询 - 删除 LlmSessionCompletionObserver（Task.Delay(50ms) 轮询 + committed 后 503 伪失败）。 - ResponsesCommandFacade.RecordCompletionAndReadAsync -> RecordCompletionAsync：持久化后直接用 已持有的 completion 构造 snapshot 返回，不再轮询 read model。 - 新增 ResponsesCompletionStage（Committed / ReadModelObserved）区分 ACK 阶段；already-resolved tool-result 路径返回 Committed，不冒充 read-model observed。 - 测试：ResponsesCommandFacadeTests 覆盖 record 成功即返回 Committed（无轮询、无 503 伪失败）。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…kflow - 新增低层抽象 port ISkillWorkflowMountPort(+ NoOp 默认实现)于 Aevatar.AI.ToolProviders.Skills； 平台层 SkillWorkflowMountAdapter(GAgentService.Infrastructure)实现之，经 IScopeWorkflowCommandPort.UpsertAsync 将 skill 抽出的 inline WorkflowYamls 注册进调用者 scope。 - use_skill 增加 mount_workflows 参数(默认 false，不破坏现有行为)；仅 loaded && mount_workflows 时挂载； scope_id / nyxid token 缺失时安全降级(missing_scope / missing_identity)，不抛未受控异常。 - 复用既有 SkillWorkflowExtractor 抽取结果与 use_skill live/per-request/identity-scoped 语义，不引入进程内 catalog。 - 保留 #1686 的 skill_load typed envelope，新增 workflow_mount 字段。 - 测试：SkillWorkflowMountAdapterTests + use_skill 挂载路径 + DI 注册。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
… superseded dev advanced 41 commits since feature/integrate branched, including its own parallel refactors that supersede two pieces of this branch's work: - #1688 (constrain ToLegacyMetadata): dev removed ToLegacyMetadata / CurrentMetadata / LegacyMetadata entirely (issues #1621 FromMetadata-external-only + #1574 metadata-scrub-only). #1688's goal is already achieved by dev; dropped (took dev's versions of AgentToolExecutionContext / AgentToolRequestContext / AIGAgentBase / mapper test / guard). - #554 (suspension enum): dev chose a different design — kept suspension_type as string and added typed WorkflowToolApprovalSuspension tool_approval (issue #1668). Dropped #554's enum (took dev's proto + workflow suspension files; removed dead WorkflowSuspensionTypes.cs). Kept + reconciled onto current dev: PR #1691, and issues #1685 / #1686 / #1687 / #1695. Reconciliation fixes: LocalSkillCatalogTests uses dev's typed AgentToolRequestContext.Current; SkillWorkflowMountAdapter(+test) updated to dev's flat ScopeWorkflowUpsertResult shape. 验证：dotnet build aevatar.slnx 0 error；architecture + test_stability guards 通过； AI.Tests 763 / Ornn.Tests 41 / GAgentService.Tests 690 / CQRS.Core.Tests 51 / Workflow.Host.Api.Tests 417 / Workflow.Application.Tests 206 / Hosting.Tests 241 全通过。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This was referenced Jun 2, 2026
- 新增 ExecutionActivityPublisherHook（IGAgentExecutionHook）：start/end/error 把 hook context 发布到
scope-keyed 流 aevatar://scopes/{scopeId}/execution-events（typed ExecutionActivityEvent protobuf）；
bounded channel + DropWrite + 后台 drain，非阻塞、失败不影响 handler；与 ExecutionTraceHook 并存。
- scopeId 从权威 inbound EventEnvelope 解析（typed scope 字段 + 窄 descriptor fallback）；解析不出则丢弃，避免跨租户泄漏。
- 新增 SSE 端点 GET /api/scopes/{scopeId}/execution-events：AevatarScopeAccessGuard 鉴权 + scope 过滤 +
evt.ScopeId 双校验；复用 host SSE 约定；只发 handler.started/completed/failed，不重复 run-output。
- 测试：scope 内 handler → 帧齐全；跨 scope 隔离；failed 带 error；非阻塞背压。
验证：build 0 error；ScopeGAgentEndpointsTests 33 通过。
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Design for forking a new WorkflowRunGAgent from a chosen step with seeded intermediate state after editing the workflow YAML. One fork primitive + two drivers (manual fork / auto on_failure retry). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ery (#1864) - proto: WorkflowRunResumeSeed, WorkflowRunForkRequestedEvent, StartWorkflowEvent.resume_seed - IWorkflowRunSeedQueryPort + WorkflowRunResumeSeedView (Application.Abstractions) - read-model seed query (no actor side-read/replay): projection transport fields 18-22, projector materialization, WorkflowRunResumeSeedReadModelMapper, WorkflowRunSeedQueryPort - tests: seed mapper, query port, proto coverage (build + 692 workflow tests + guards green) Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…#1866) - HandleStartWorkflowAsync honors StartWorkflowEvent.resume_seed: hydrate variables (precedence seed -> input -> start params), start at resume_seed.start_at_step_id instead of first step, loud-fail when the seeded step is absent, dispatch with seeded "input" when present - does NOT reuse WorkflowResumedEvent (suspension-resume is a distinct path) - tests: valid seed dispatch+hydrate, missing step failure, precedence, no-seed regression (build + Core.Tests 431 + guards green) Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…1868) - WorkflowForkRunCommand + IWorkflowForkRunService/WorkflowForkRunService (Application/RunForks): query seed via IWorkflowRunSeedQueryPort -> validate source terminal + YAML compiles + start step exists (structured errors) -> CreateRunAsync(chosen YAML) -> dispatch WorkflowChatRequestEvent carrying resume_seed; compensates orphan actors on dispatch failure - proto: WorkflowChatRequestEvent.resume_seed = 11; app model WorkflowChatRunResumeSeed + WorkflowChatRunRequest.ResumeSeed; envelope factory maps app model -> proto - WorkflowRunGAgent.HandleChatRequest forwards ResumeSeed into self-published StartWorkflowEvent and resolves execution input from seed["input"] (consistent Variables["input"]) - seed rides the execute trigger, not the bind; new run never side-reads old run - tests: terminal/missing/yaml-compile errors, override precedence, happy-path dispatch, envelope (build + Core 431 + Application 262 + guards green) Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…point (#1872) - WorkflowForkRunInput + HandleForkRun in WorkflowCapabilityEndpoints: validate required (400) -> WorkflowForkRunCommand -> IWorkflowForkRunService.ForkAsync -> map result: accepted 202 + Location (.../workflow-actors/{newRunId}/current-state) + honest body; SourceRunNotFound 404, NotTerminal 409, InvalidYaml/StartStepNotFound 422, create/dispatch 502 - ACK honest (only fork-dispatch admission, not applied continuation) - tests: accepted+location+mapped dispatch, missing-required 400, error mapping (build + Host.Api.Tests 465 + guards green) Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…nator (#1874) - WorkflowParser/WorkflowDefinition parse top-level on_failure {action: fork_from_failed_step, max_attempts} (run-level; distinct from step-level retry/on_error) - WorkflowRunGAgent emits committed WorkflowRunForkRequestedEvent on terminal failure when policy matches and fork_attempt < max_attempts (decision inside the actor event handler) - attempt propagation bounds the chain: WorkflowRunForkRequestedEvent.attempt -> command -> seed -> proto -> WorkflowRunExecutionStartedEvent.attempt -> WorkflowRunState.fork_attempt - WorkflowRunForkCoordinator (ICommittedStatePublicationHook) consumes the committed event and calls IWorkflowForkRunService.ForkAsync(InlineYaml:null) -> reuse source YAML; no persistent-definition dep - kernel CleanupRunAsync preserves terminal facts so a terminal run's seed stays readable (required for forking any terminal run; necessary fix the earlier units missed) - tests: on_failure parse, failed+policy emit, no-policy/attempt>=max no-emit, coordinator->ForkAsync (build + Core 436 + Application 263 + guards green; workflow integration tests clean) Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…scheduled skill runs (#1849) Adds SkillRunnerOutputFormat enum threaded through skill_runner.proto (SkillRunnerOutboundConfig + InitializeSkillRunnerCommand), the scheduled_agent_creator schema + mapper, SkillRunnerGAgent output branching (FEISHU_DOC forces a Feishu doc, TEXT forces chunked text, AUTO keeps length-based behavior), and the user_agent_catalog projection + readers. Closes #1847. Verified: affected projects build; ChannelRuntime.Tests 1085/1085 pass; stability/query/projection guards pass. Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…apper + workflow typed lift (#1869) * feat(#1861): typed InteractionSpec contract (Foundation) + InteractionSpecMapper + workflow typed lift (design-consensus structural:Foundation) * fix(#1861): 删除 dead HasVisibleContent + 补 WorkflowParser root-level/inline presentation 分支测试 reviewer r1: tests reject(net-new 分支缺测)+ quality comment(dead public API)。 按 controller 决策(删除优先)删除无调用方的 InteractionSpecMapper.HasVisibleContent， 并补 root-level interaction_spec source 与 inline presentation lifting 两个 WorkflowParser 行为测试。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
…tion (#1845) (#1850) * feat(nyxid-chat): system-prompt playbook for autonomous long-running task automation Adds a 'Long-running task automation playbook' to the bot system-prompt guiding it to: recognize recurring-task requests, author a runnable active-playbook skill (mirroring deadline-monitor: collect via nyxid_proxy->api-github, analyze, deliver to Lark), publish via ornn_publish_skill, negotiate frequency/time/delivery/output via an interactive card, then schedule 24x7 via scheduled_agent_creator (+ agent_delivery_targets rebind), with failure/rejection recovery. Adds a regression test asserting the playbook + tool references. Closes #1845. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(#1845): 删除生产 prompt 里 deadline-monitor 具体技能名硬编码，改泛化 discovery 表述 + 同步测试 architect review r1 reject：system-prompt.md 在生产 prompt 里硬编码具体 skill/模板名 deadline-monitor，违反 CLAUDE.md:22。 改为基于已加载 skill metadata/instructions 的泛化监控/digest 表述；测试改断言通用行为并加 deadline-monitor 不出现的负断言。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… Lark boundary (#1873) * feat(#1863): 出站交互 notify primitive — typed notification event chain + Lark boundary design-consensus structural（issue #1863）：新 notify primitive 走独立 typed 出站事件链，不挂起、不复用 HITL suspension。 - NotifyModule（校验 delivery_target_id + interaction/interaction_template 恰一 → 发 WorkflowInteractionNotificationEvent → StepCompletedEvent(accepted)，绝不发 WorkflowSuspendedEvent） - WorkflowInteractionNotificationEvent proto（oneof InteractionSpec/InteractionTemplateSpec）+ Foundation InteractionTemplateSpec - WorkflowInteractionNotificationProjector → IChannelInteractionNotificationPort（+ Null 实现） - FeishuCardNotificationPort + 抽出 FeishuCardOutboundMessageSender 与 HITL 共享 - architecture_guards.sh FI-006：Workflow Core 禁引用 channel 渲染 - ACK 诚实：accepted=event published，不暗示 Lark delivered Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fix(#1863): delivery_target_id 强类型化 + 补 dispatch/guard 测试 + 删 dead Annotations (review r1) - architect reject：delivery_target_id 在通用 bag → 加 typed WorkflowStepParameters.delivery_target_id=6， WorkflowParser lift 到 StepPresentation.DeliveryTargetId、kernel 求值写入、NotifyModule 只读 typed；删 camelCase alias - quality reject：删除无 producer/consumer 的 ChannelInteractionNotificationRequest.Annotations - tests reject：补 kernel 真实 dispatch template 求值测试、parser typed-lift 断言、FI-006 guard source-regression、proto round-trip Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
…val) (#1876) * feat(#1862): HITL typed interaction consumer chain (human_input/approval) design-consensus r4 structural（issue #1862）：human_input/human_approval 携带 typed InteractionSpec， 经 WorkflowSuspendedEvent → WorkflowHumanInteractionProjector → IHumanInteractionPort/FeishuCardHumanInteractionPort 投递； 回流只消费 #1846 typed CardActionSubmission.action_kind 的 FormSubmit gate。 HITL 显式拒绝 interaction_template（仅 notify 用）。只消费 #1861/#1846/#1863 已落地 typed surface，不改 parser/step/kernel。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fix(#1862): HITL approval 决策强类型化 + 去重 HasInteractionSpec (review r1) architect reject：approval 决策从 action_id 字面推导 → 加 typed InteractionApprovalDecision + InteractionAction.approval_decision（Foundation 扩展，architect 要求 typed modeling）， FeishuCardHumanInteractionPort 从 typed payload 构造 WorkflowResumeActionPayload.Approved，删除 action_id==approve/reject 控制映射； WorkflowParser 解析 typed approval_decision、InteractionSpecMapper 映射；action_id 只作 UI 标识。 quality comment：3 处重复 HasInteractionSpec 改用 canonical StepPresentation.HasInteractionSpec。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
…ow run 失败恢复 batch 汇总合并 eanzhao 两天内 issues 的实现（经 consensus-rnd loop 评审）: - 交互协议: #1861 Foundation typed InteractionSpec + #1846 reply_with_interaction 原语 + #1863 出站 notify + #1862 HITL typed consumer - workflow run 失败恢复: #1855 #1856 #1857 #1858 #1859 (epic #1854) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Reserve implementation PR for issue #1885 * Fix scheduled agent creator approval mode * Add scheduled agent prompt approval regression test
…uled creator on relay path
Surfaced by a real Lark automation test ("每天5点发我 github deadline 概要").
Skill discovery: the long-running automation playbook authored a new skill without searching first, creating a duplicate deadline-monitor instead of reusing the published deadline-monitor-digest-payload-builder. Add a 'search Ornn first, reuse via use_skill' step before authoring.
Scheduled scope: BuildLlmReplyRequestAsync never threaded the bot registration scope into the deferred LLM-reply tool context, so scheduled_agent_creator failed with scope_id_unavailable. Set Caller/Channel/ExternalMetadata mirroring the direct-reply BuildAgentBuilderToolContext; ToToolContext only overlays creds/routing so the typed scope survives to tool execution.
Add a regression test asserting the LLM-reply tool context carries the scope; update one stale assertion that pinned the old scope-absent behavior.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…und Lark bot After scope threading, scheduled_agent_creator next failed with lark_outbound_provider_slug_unavailable: the inbound chat path wrote InboundChannelBotProxySlug but never LarkOutboundProxySlug, which the mapper requires for the primary outbound provider. Set it to inboundEvent.NyxProviderSlug so a chat-triggered scheduled task delivers via the same Lark bot that received the message; a distinct outbound provider stays expressible via agent_delivery_targets. Extend the LLM-reply tool-context regression test to assert the outbound provider slug is exposed. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Squash merge release rollup after resolving conflicts with feature/integrate.
… DM 为空 (#1916) * fix(nyxid-chat): relay registration 按 canonical scope 选择，修复 /agents 在 DM 为空 ChannelConversationTurnRunner.ResolveRegistrationAsync 改用 ListByNyxAgentApiKeyIdAsync： 有 TransportExtras.NyxRegistrationScopeId 时只接受 ScopeId 精确匹配的候选，否则 fail closed； 无 canonical scope 时只允许折叠到唯一非空 ScopeId（0/全空/多个 distinct 均 fail closed）； bot-id fallback 仅在 API-key 候选为空时参与，并校验 canonical scope 与 API key。 删除 GetByNyxAgentApiKeyIdAsync 单结果与 QueryAllAsync first-hit bounded scan， 使 DM 与群聊解析到同一 registration，strict OwnerScope 过滤不再漏掉用户自建的 scheduled agent。 保留 /agents 的 strict OwnerScope 四元组隔离（#466），不改 NyxID/chrono-* 外部仓库。 Closes #1914 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * test(nyxid-chat): 补 registration scope 选择的两个 fail-closed 分支测试 - identity-query port 缺失但带 NyxAgentApiKeyId 时应返回 registration_not_found，不回落 bot-id - API-key 候选为空且 bot-id fallback 的 ScopeId 与 canonical 不一致时应拒绝 补强 review r1 中 tests reviewer 指出的覆盖缺口（PR #1916）。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* Reserve implementation PR for issue #1886 (#1888) * 强化 direct Responses 本地工具失败处理 (#1895) * 实现 issue #1886 * Add user skill cancellation coverage * 修复 workflow fork run 命令分发与 scope 继承 (#1887) * Reserve implementation PR for issue #1881 * Remove spawn-codex.sh script: deprecated execution wrapper for codex-refactor-loop skill * docs: 记录 aevatar 为 consensus-rnd:codex-refactor-loop dogfood 仓库 - 新增 Dogfood 段：本仓库是该 skill 的 dogfood host - skill 自身 bug/使用问题及时自动提到上游 ChronoAIProject/consensus-rnd - daemon 只跑最新版,本仓库历史版本进程即停(仅约束本仓库) - host.env 为运行时注入点,缺失即停不臆造 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * 实现 issue #1605 (#1844) * 收敛团队创建页路由契约 (#1843) * 实现 issue #1770 * 实现 issue #1770 * PR #1843 remote-ci-fix output * PR #1843 review-fix output * PR #1843 remote-ci-fix output * PR #1843 review-fix output * Improve team member entry actions (#1867) * Improve team member entry actions * Avoid duplicate team members actions * PR #1887 review-fix output * PR #1887 review-fix output * 实现 issue #1881 * PR #1887 review-fix output * PR #1887 review-fix output * Remove unused workflow resume seed name * Use shared command pipeline for workflow fork dispatch --------- Co-authored-by: Loning <aloning@gmail.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Co-authored-by: potter-sun <82434926+potter-sun@users.noreply.github.com> * 收紧 Aevatar 运行观察工具契约 (#1908) * 实现 issue #1894 * PR #1908 review-fix output * Fix Lark card finalize delivery truth (#1910) * 实现 issue #1884 * PR #1910 review-fix output * 预留 issue #1883 实现分支 (#1889) * Reserve implementation PR for issue #1883 * 实现 issue #1883 * PR #1889 review-fix output * PR #1889 review-fix output * PR #1889 review-fix output * PR #1889 review-fix output --------- Co-authored-by: Loning <aloning@gmail.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Co-authored-by: potter-sun <82434926+potter-sun@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
摘要
集成分支
feature/integrate：把 eanzhao 的可用 PR + 一批架构/能力 issue 合到一条分支，已 reconcile 到当前dev。包含的工作
PR 整合
Forge 自我扩展 / runs 可观测 / OpenAI 兼容 issue
use_skill mount_workflows：把 skill 的assets/*.yaml挂载为 scope 可调用 workflowUsageEvent（usage=16），workflow/nyxid/draft/script 路径 emit token usage/api/scopes/{scopeId}/execution-events）/v1/chat/completionsobserve/stream-back（含 bounded observe 超时防挂死）因 dev 已超越而弃用（同步时丢弃 + 关闭）
tool_approval（[refactor-design] C1 workflow tool_call ApprovalPending 不应成为终态失败 #1668）AgentToolExecutionContext.ToLegacyMetadata()使用边界，防止内部控制字段扩散到 metadata #1688（收窄 ToLegacyMetadata）→ dev 直接删除了 ToLegacyMetadata（[refactor-design][1595-first] demote-FromMetadata-ToLegacyMetadata-to-external-only #1621/[refactor-design] #1569-later: legacy-tool-metadata-ingress-boundary-design-pending #1574）验证（本地）
dotnet build aevatar.slnx --nologo：0 errorarchitecture_guards.sh/test_stability_guards.sh：通过Closes #1685
Closes #1686
Closes #1687
Closes #1695
Closes #1699
Closes #1700
Closes #1701
Closes #1703
Closes #1704