feat(interaction): real mid-turn steering with applied reporting

[renanalmd]

Paired: GitHub #497 ↔ GitCode !2305

What type of PR is this?

/kind feature

What does this PR do / why do we need it:

Wire send_input so hosts can steer an active round without stealing the output lease, refuse steers after the round ends, and report what reached model context.

• SendInputResult / InputDisposition on every send_input path
• STEER with no active round rejects (no_active_round); expected_round_id rejects round_mismatch when the client races past turn end
• Drop undrained steers at round teardown so they cannot bleed into the next turn; emit steer.applied with dropped ids
• Steering envelopes carry optional request ids; rails can bound a drain
• Team leader and member steers reach the live runtime; applied/dropped surfaces for hosts
• Unit coverage for interaction, team harness, and applied reporting

Hosts need a truthful mid-turn path. A bare success flag hid follow-up downgrades and end-of-turn races. This contract is the dependency for the JiuwenSwarm chat.steer PR.

Which issue(s) this PR fixes:

Fixes #2748

What scenarios were tested, and what were the verification results（Function, performance, reliability, etc.）：

Function

• pytest tests/unit_tests/harness/test_deep_agent_interaction.py — steer queue, expected_round_id, no_active_round, undrained drop at round end (steer-related cases passed)
• pytest tests/unit_tests/harness/test_steering_applied.py — applied/dropped reporting
• Team: test_steer.py, test_team_harness_steer.py, test_steer_leader.py

Reliability

• Steer with matching expected_round_id → steer_queued
• Wrong or stale expected_round_id → round_mismatch (does not fall through to follow-up)
• Idle steer with nothing open → no_active_round
• Leftover queue at round end → steer.applied with empty applied and dropped ids

Contract smoke (publish install)

python -c "from openjiuwen.harness.schema.interaction import SendInputResult; print(SendInputResult)"
python -c "from openjiuwen.core.runner.runner import Runner; assert hasattr(Runner, 'steer_agent_team')"

Performance: not a target of this change.

Security: steer does not take or replace the output lease.

Self-checklist:（Please check carefully,and mark an x in the [] brackets. We will review your completion status.）

• • Design: Has the solution corresponding to the PR been reviewed by the Maintainer, and have all review comments been replied to and revised
• • Test: Has the code in the PR been fully covered by UT/ST test cases, and have the newly added test cases been uploaded to the repository along with this PR or already uploaded.
• • Verification: Does the PR description contains a detailed description of the verification results regarding the achievement of the expected goals for the Feature, Refactor, and Bugfix to this PR.
• • Interface: Does it involve changes to external interfaces? The corresponding changes have been approved by the interface review organization, and the annotation information for the API has been correctly refreshed.
• • Document: Does it involve modifications to the official website documentation? If so, please submit the materials to the Doc repository in a timely manner.

Special notes for reviewers

• Public API: SendInputResult replaces a None return from send_input. Hosts must read accepted / disposition / reason.
• Publish this package before JiuwenSwarm bumps its pin.

[openjiuwen-collaboration-bot]

head_sha: b52d8f00cc02046bc7593e12131895264230e212

变更摘要

此 PR 为核心交互层引入了真实的**中途转向（mid-turn steering）**机制，并附加了完整的应用报告。核心变更在于：send_input 现在返回 SendInputResult（包含 accepted/disposition/reason），STEER 模式在没有活跃轮次时会被拒绝（而非静默降级为新轮次），转向文本在轮次结束时会被排空并报告，且带请求 ID 的 SteeringInput 信封贯穿整个队列链路，使主机能够区分“仍在处理”和“已被 rail 丢弃”。

主要改动

• DeepAgent.send_input 返回 SendInputResult 并拒绝无效转向：send_input 从 -> None 改为 -> SendInputResult，新增 InputDisposition 枚举（STEER_QUEUED/FOLLOW_UP_QUEUED/TURN_QUEUED/RESUME_QUEUED/REJECTED）。当 mode=STEER 且无活跃轮次时返回 rejected("no_active_round")，而非像旧代码那样降级为全新轮次；带 expected_round_id 的转向若不匹配则返回 rejected("round_mismatch")。

• 新增 SteeringInput 信封与全链路 ID 传递：loop_queues.py 新增 SteeringInput(text, id) 数据类，push_steer、enqueue_steer 和 drain_steering 均从裸字符串升级为支持该信封。DeepAgent.send_input 在 STEER 路径中将 request.request_id 注入 SteeringInput，确保 STEER_APPLIED 事件的 dropped 列表能正确构建。

• 新增转向应用报告与轮次结束时排空：ReActAgent._report_steering_applied 在 rails 处理后报告哪些转向文本到达了模型上下文，哪些被丢弃；DeepAgent._drop_undrained_steering 在轮次结束时排空残留转向队列并发出 steer.applied 事件（applied=[]）。NativeHarness._emit_interaction_event 将事件写入 output_queue，修复了 Team 客户端此前收不到 steer.applied 的问题。

• 新增 SupportsRoundSteering 协议与 TeamAgent.steer/TeamHarness.steer_round：在 member_runtime.py 中从 MemberRuntime 分离出 SupportsRoundSteering 协议（声明 steer_round 方法），使 ExternalCliRuntime 不会因此失去 MemberRuntime 符合性。TeamAgent.steer 通过 isinstance 检查路由到运行时，不支持时抛出 NotImplementedError；TeamHarness 新增 steer_round 和 active_round 属性作为转发面。

• TeamRuntimeManager.steer_leader 与 Runner.steer_agent_team：新增独立的领导者转向入口，通过 has_in_flight_round 快速检查、interact_gate 控流、以及 expected_round_id 防竞态，返回 DeliverResult 并区分 not_active/no_active_round/gate_closed/unsupported_runtime。

[openjiuwen-collaboration-bot]

head_sha: 3ca8ed02960a3bf297095973371bec838da10655

🤖 AI Code Review

⚠️ 审查未能完成

本次代码审查未能完成，可能是临时性故障。

你可以在评论区输入 /ai review 重新发起审查。若多次失败，请联系管理员并附上下方追踪 ID。

追踪 ID: task-179145-run-172043

[openjiuwen-collaboration-bot]

head_sha: b52d8f00cc02046bc7593e12131895264230e212

任务名称	结果	日志操作
静态检查	❌FAILED	点此跳转
防投毒检查	✅SUCCESS	点此跳转
开源合规检查	✅SUCCESS	点此跳转
UT测试	❌FAILED	点此跳转
ST测试	N/A	N/A
build 编译包	N/A	N/A
ruff codecheck	❌FAILED	点此跳转

[openjiuwen-collaboration-bot]

head_sha: 4795ece3f9157af3693672993bdaeda948f3dd90

任务名称	结果	日志操作
静态检查	✅SUCCESS	点此跳转
防投毒检查	✅SUCCESS	点此跳转
开源合规检查	✅SUCCESS	点此跳转
UT测试	❌FAILED	点此跳转
ST测试	N/A	N/A
build 编译包	N/A	N/A
ruff codecheck	❌FAILED	点此跳转

[openjiuwen-collaboration-bot]

head_sha: 3ca8ed02960a3bf297095973371bec838da10655

任务名称	结果	日志操作
静态检查	✅SUCCESS	点此跳转
防投毒检查	✅SUCCESS	点此跳转
开源合规检查	✅SUCCESS	点此跳转
UT测试	✅SUCCESS	点此跳转
ST测试	N/A	N/A
build 编译包	N/A	N/A
ruff codecheck	❌FAILED	点此跳转