Repository navigation
Description
Current state: Accepted and complete for the bounded T38 delivery. Implementation #475 merged as 0311393c; the authorized Harness owner accepted completion and closure. Production evaluation remains tracked in #429, #430 and #434.
Goal
Track the Harness retention and sealing slice required by #371 step 4 and D023-3 / V33 / T38. The Apps adapter in #446 hands over process-local bytes, actual fixed Knowledge pins, selection/omission metadata, a proposed location and SHA-256. Harness owns retention and the committed delivery reference.
Accepted scope and acceptance
- The Harness generation orchestration consumes the injected
ContextProvider.collect(ContextRequest)before publishing a generation manifest. The implementation identifies the concrete entry point and Harness capture/commit route in its PR and exercises both through that entry point. - Harness seals
DeliveredContextwith the actual committed locator, which may differ from the draft proposal. ItsArtifactRef.sha256equals the draft digest and itssize_bytesequals the retained byte count. The accepted fixed media type isapplication/octet-stream; Harness owns this field. - Stage retention by attempt. When the same round is retried, an unsuccessful attempt's staged bytes cannot satisfy a committed read, and an existing committed delivery cannot be overwritten. If an already committed immutable object is reused, verify its exact digest and length. The draft location is a relative hint.
- T38-N1 injects two failures: before writing bytes and after staging the bytes but before commit. For each, verify no valid
DeliveredContext.artifact, no entry inGenerationManifest.context_refs, and no new sealed generation manifest for the failed delivery. - T38-P1 observes the order of retention, sealing and manifest publication. Verify the manifest's
context_refsincludes the committedDeliveredContext.artifact. Read the retained bytes in a new process under the same configuration and compare the exact bytes, digest and fixed Knowledge pins. - Exercise the actual Apps adapter with a dependency check or write-failing spy around Apps collection. The writer of record is Harness. The process-local draft stays outside the JSON schema; the committed
DeliveredContextandGenerationManifestuse the existing schema.
Shared D023-3 contract · T38 acceptance · Amy's review
Ownership and status
Harness owner: @Dong1017; bounded implementation delegated to @zengchuang-hw in #475. Completion and closure explicitly accepted by Dong1017's authorized delegate in the acceptance record. See the closing summary for artifacts, verification and limits. #371 and the three production evaluation dependencies remain open.
Activity
amy-why-3459 commentedon Nov 15, 2026
Thanks for splitting this out. It keeps the D023-3 boundary clear: #446 produces the process-local draft, and Harness owns retention and sealing. Some points that would make the acceptance easier to verify:
ContextProvider.collect(ContextRequest)and the capture/commit route it uses to retain the bytes (an existing artifact capture path, or a new one). That way the implementation and T38 test have a fixed entry point.ArtifactRefrequireslocation,sha256,size_bytesandmedia_type. The draft has the bytes, the digest and a proposed location, but no media type. Please state who setsmedia_type(I'd expect Harness to fix one value for delivered context) and thatsize_bytesislen(content)as retained. Please also say whether the committed locator may differ fromintended_location, and thatDeliveredContext.artifact.sha256must equal the draft'ssha256.round_id(context/<sha256(round_id)>/delivered.binin Adapt Knowledge context to a Harness delivery draft #446). Please define what happens when a round is retried after a failed retention: overwrite, a per-attempt location, or rejection. This matters for T38-N1, so leftover bytes from the failed attempt can't be picked up later.DeliveredContext, no sealed generation manifest, and nocontext_refsentry for that delivery.context_refspoint at the sealed delivery and that the manifest is written strictly after sealing.The owner is still open (Proposed, waiting for @Dong1017). Once that's confirmed, I'm happy to review the Harness slice against T38-N1/P1.
zengchuang-hw commentedon Nov 15, 2026
@amy-why-3459 the issue now incorporates your six checks. It names the generation orchestration seam and asks Harness to record its exact entry point and capture/commit route. The proposal includes a fixed media type, length and digest equality, attempt-scoped staging, both pre-commit failure points, the manifest ordering and context refs, and a check on Apps' read-only dependency boundary. The concrete route and the accountable implementer are pending @Dong1017's acceptance.
amy-why-3459 commentedon Nov 15, 2026
Thanks, I checked the updated "Proposed acceptance" against my six points (6889009504). All of them are covered:
ArtifactReffields: committed locator may differ from the hint,sha256equals the draft digest,size_bytesequals the retained length, and Harness ownsmedia_type, proposed asapplication/octet-stream(item 2).artifact,context_refsand the manifest (item 4). T38-P1 checks ordering and thatcontext_refscontains the committed artifact (item 5). A spy or dependency check enforces the writer boundary (item 6).Nothing else from me on the acceptance text. What's still open is @Dong1017's confirmation of the Harness entry point, the capture/commit route, the media type and the implementer.
Dong1017 commentedon Nov 15, 2026
Decision — T38 Harness entry / route / media / implementer (Dong-bot as @Dong1017 delegate)
Recorded 2026-11-15. Authority: Dong-bot decision authority for
@Dong1017accept/decide. Thread basis: Amy cleared proposed acceptance (6889114513); D023-3 / V33 / T38; Apps draft producer #446; Harness owner precedent #429.Accepted — proposed acceptance text
Accept this issue’s current “Proposed acceptance” as the bounded Harness retention/seal delivery for #371 step 4 / D023-3 / T38. Amy’s six checks are covered; no further acceptance-text changes required from steward.
Confirmed — concrete ownership
GenerationManifest(V24/V33). It must call the injectedContextProvider.collect(ContextRequest)before sealingDeliveredContextand before writing any sealed generation manifest. Notexecute(), and not the legacy optimization-sessioncapture_contexthelper. The implementation PR must name the concrete module/symbol for this controller and exercise T38-N1/P1 through that entry only.Claim.commitsoArtifactRefbecomes valid only with the commit (same writer family as H02 capture→commit; writer of record = H). Draftintended_locationis a relative hint only; the committed locator may differ. Failed-attempt staged bytes must not satisfy a committed read; an existing committed delivery must not be overwritten; immutable reuse requires exact digest+length verify. Capture-route identity for producer metadata:delivered-context-retention. Apps must not write the H RunStore.ArtifactRef.media_type. Fixed value for delivered context:application/octet-stream.sha256equals the draft digest;size_bytesequals the retained byte count (len(content)).Scope note
#447 is the Harness delivery issue for T38 retention/seal. #446 remains the Apps draft producer only; it does not cover T38-N1/P1 production retention. No other open issue already covers both T38 cases for Harness.
Status
Flipping
status:proposed→status:acceptedfor this bounded delivery. Apps can wire the real Knowledge draft + joint CPU tests once the named entry/route land.zengchuang-hw commentedon Nov 16, 2026
@Dong1017,Apps 的
CandidateProvider.plan_generation已随 #465 合入,collect_candidate(record, artifacts=...)已提交 #469,可从普通 CPU 执行的已提交记录读取该 attempt 的固定 manifest 和产物。回归测试直接提供 V24 需要的generation_plan输入,Harness 生产编排里还需要对接 #447 的上下文留存和封存。请把 Harness 生成控制器的实现 PR 或分支链接贴到这里。Apps 可沿用 #469 的用例,对接实际
ContextProvider及CandidateProvider,补齐 T38 的提交前失败和新进程重读验证。Dong1017 commentedon Nov 16, 2026
Posted by an authorized steward agent on behalf of @Dong1017 (decision authority per standing Dong-bot / steward delegation; hosted Dong-bot is on
usage_limit).Status — Harness generation-controller PR for T38
Thanks @zengchuang-hw — noted that Apps
CandidateProvider.plan_generationis onmainvia #465 andcollect_candidatehas landed via #469 (squash-merged this round).Harness production retention/seal PR: there is no open Harness generation-controller PR/branch for #447 yet. Accountable delivery remains with the Harness owner (@Dong1017 / S2) per the prior decision (688920… thread). When that PR opens, its URL will be posted here; until then Apps can keep using the #469 test pattern (V24 sealed
generation_plansupplied in the CPU path) and must not treat Apps collect as T38-N1/P1 production retention.Binding reminder (unchanged): entry = Harness generation orchestration that publishes
GenerationManifestand calls injectedContextProvider.collectbefore sealingDeliveredContext/ writing the sealed generation manifest; capture/commit = RunStore attempt-scoped stage thenClaim.commit; media type =application/octet-stream.zengchuang-hw commentedon Nov 16, 2026
@Dong1017,#465 和 #469 的 Apps 端口已在主分支,我这边开始验证 #447 的真实 CPU 接入。当前 RunStore 有一个顺序卡点。
Claim.commit要先claim(attempt, request_digest),而 V24 要求先提交交付上下文、写入并绑定generation_plan,再用最终 spec 计算request_digest。此外,通用对象路径在 SQLite 记录提交失败后可能留有未引用的文件,不能直接当作 T38 的已提交读取凭据。我建议在 Harness RunStore 增加按 attempt 和角色提交的前置留存事务。只有事务提交后的槽位才能经固定 Ref 读取。控制器依次提交交付字节和固定 pins,生成
DeliveredContext,写入并绑定 manifest,再走现有 execute 登记。普通执行记录不预登记、不改摘要。请确认这个存储接口选择,并明确这段是否由我提交实现 PR。如已有分支,请贴链接,我会接着补实际 AppsContextProvider、T38 两个失败点和跨进程 CPU 用例。#429、#430、#434 如果已有可联调的 CPU 分支,也请一并贴出。本地实现和测试已放到 draft PR #475。T38 的两个失败点、跨进程重读实际 bytes 和固定 pins、以及使用已封存 spec 的真实普通 CPU 执行和候选回收均通过。请优先检视前置留存事务是否符合 Harness 的存储边界。
Dong1017 commentedon Nov 16, 2026
Posted by an authorized steward agent on behalf of @Dong1017 (decision authority per standing Dong-bot / steward delegation; hosted Dong-bot is on
usage_limit).Decision — T38 pre-registration retention transaction (#475)
@zengchuang-hw — confirmed for Harness storage boundary:
ContextProviderwiring and the two T38 failure points + cross-process CPU cases there.generation_inputsneeds aBEFORE DELETEimmutability trigger (and append-only contract wording) — formal Request changes on headcc9b2206: Seal Harness generation inputs before execution registration #475 (review). david6666666's retry/reuse note remains in scope for the same PR.No change to the earlier T38 media / route decisions on this Issue.
zengchuang-hw commentedon Nov 17, 2026
@Dong1017,已记录你对 #475 前置留存接口的确认。我会先补齐数据库的追加写保护,以及同一请求重试时复用已提交交付的逻辑,然后把 #475 转正式检视。
请同时明确 CPU 生产联调的三个依赖:#429 Harness 评测执行、#430 Benchmark prepare、#434 Benchmark 离线 assess。已有实现请贴 PR 或分支链接,尚未开工请给出预计可联调时间。三个端口接通后,我会跑 #371 的固定记录评测闭环。
zengchuang-hw commentedon Nov 19, 2026
@Dong1017 更新 #371 的生产联调状态:#475 已合入
0311393c62330c041a01238dc86823b80e020561,实际入口为inferrsi.harness.generation.prepare_generation。它按已接受的顺序提交交付 bytes/pins、封存 manifest、绑定最终 spec,并覆盖两处回滚、原请求恢复、跨进程固定读取、CPU 普通执行及候选回收。请确认这个交付是否满足 #447 的 T38 限定验收,并明确可否完成/关闭该子任务;尚有缺口请逐项指出,我接着处理。生产评测仍不能视为完成:请同时贴 #429、#430、#434 的实现 PR/联调分支或预计可联调时间,三者已于 11 月 15 日接受并指定由你负责限定交付。
Apps 当前按 #490 整理目录,storage/handoff 已合入,workflows #517 必需 CI 已全部通过,正在等第二份有效审批;接口签名与固定引用契约保持原有边界。三项生产端口就绪后再执行共享 CPU T36/T37/T39/T40,现有普通执行回归不替代该生产闭环。
Dong1017 commentedon Nov 19, 2026
Posted by Dong-bot, an automated agent acting on behalf of @Dong1017 (always-on-bot, model claude-sonnet-5-5).
Decision: accepted. #475 (
0311393c, entryinferrsi.harness.generation.prepare_generation) satisfies the bounded T38 Harness retention/seal acceptance in this issue. #447 can be completed and closed. (Dong-bot, authorized delegate. I checked your summary against the accepted clauses; I have not read the merged diff or run the tests.)Clauses covered by your summary
generation_inputsBEFORE DELETEtrigger and the retry/reuse handling) were required before merge. Per the 11-16 decision, the merge needed dual Approve and passing required checks.Not stated in your summary. Please add one line each in the closing comment or the #475 reference, citing the test or symbol. These are non-blocking and don't hold the close.
media_type=application/octet-stream,sha256equal to the draft digest, andsize_bytesequal tolen(content).ContextProviderpath is exercised, with a check that Apps makes no RunStore writes. The Harness slice is accepted without this. If the Apps-side wiring is still pending, track it on the Apps side or in task: Apps 应用层详细设计与迁移计划(CLI / Python / Plugin) #371, not here.Production evaluation is not complete. #429, #430 and #434 stay open. This thread has no PR or branch link for any of them, and I can't supply an ETA from it. Their accountable owner is still the Harness owner (@Dong1017 / S2), as accepted on 11-15. The shared CPU T36/T37/T39/T40 run should wait for all three ports. Existing ordinary-execute regressions don't count as that closure. The owner needs to post PR/branch links or dates for these three on #371, or on their own issues. That's a separate follow-up from closing #447.
zengchuang-hw commentedon Nov 19, 2026
已验收:Harness 上下文留存与封存(T38)
flowchart LR A["Apps 提供上下文 bytes 与 pins"] --> B["Harness 按 attempt 和角色提交"] B --> C["封存 manifest 与最终 spec"] C --> D["原 attempt 普通 CPU 执行"] B --> E["固定引用跨进程读取"]0311393c62330c041a01238dc86823b80e020561;入口inferrsi.harness.generation.prepare_generation。由 Dong1017 的授权代理明确接受本 issue 的限定验收并同意关闭。b67e79d1再跑tests/test_harness_generation_seal.py,9 passed。覆盖两处交付回滚、不可变槽位、同请求恢复、跨进程 bytes/pins、最终 spec 绑定、普通 CPU 执行和原 attempt 重放。补充验收记录与 owner 要求的两项引用
第 2 项(字节元数据): run_store 的
_GENERATION_MEDIA、artifact_ref和_generation_ref为 delivered-context 固定application/octet-stream,从实际 bytes 计算 SHA-256 和len(data);generation._collect_delivery 先校验实际 content 摘要等于 draft.sha256,再调用 Harness 提交。这里分别引用源码责任点,不把未写出的逐字段断言冒充现有测试。第 6 项(实际 Apps 路径和写入边界):
cpu_generationfixture 与test_seal_commits_context_and_plan_before_cpu_execution注入真实 AppsContextProvider。此外在本轮候选实际执行仓库静态依赖闭包检查:19 个可达模块,0 条 Harness 依赖,排除 Apps collection 经该静态依赖到达 RunStore。复核代码:这项是依赖检查,未声称运行了独立 writer-spy。源码中的 Apps provider 只返回进程内 draft,由 Harness 执行留存和封存。owner 对 #447 的完成接受与这里的软件回归均不构成性能收益或整项 S5 的完成声明。