Repository navigation
RFC: unify Goal evolution and scalable acceptance contracts - #8752
Conversation
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
loopx-agent
left a comment
There was a problem hiding this comment.
Reviewer: model_agent · gpt-6.1-sol · OpenAI · max · runtime_reported
动机
长期运行的 Goal 收到用户纠正时,负责当前任务的 Agent 和依赖该成果的同伴都需要知道哪一版要求有效。
旧实现把所有声明放在同一个 128 KiB JSON 预算里,接近上限时新纠正无法保存。提案为已满 Goal 规定可恢复的新增操作,并为更改已有条件规定显式的版本替换、受影响租约处置和后续采用。
本次交付是可独立复审的双语设计,明确了从用户纠正到原请求收到结果的语义、验收和迁移顺序。
本 PR 的交付范围是设计。生产运行时、已安装宿主、共享存储资格和模型采用留在对应实现与验收里。
已满 Goal 的保存与恢复需要 M1 实现;联动修订、并发和真实 peer 采用仍由 M2 验证。
改动思路
原请求、当前要求、历史和证据使用各自的生命周期。提案由既有 Goal acceptance/amendment owner 校验来源、权限和版本,由既有存储提交记录。任务读取绑定版本,受影响 Agent 在下一次相关工作边界自行采用。关联信息不足时要扩读并公开缺口。
以 docs/architecture/rfcs/loopx-overall-roadmap-v0.md#r4-shared-goal-alignment-and-evolution 的 6fef4ea6b2d4a17438f51e5dee1b2fc02c1cba28 为规范基线。R4 的权限、CAS、在途处置和无关工作的连续性落在 §4、§6 与 E2/E3。R5 的完整来源、容量、回执和恢复映射到 §5、§8 与 E1/E4/E8/E9,实际资格留给 M1/M3。
具体改动
精确提交 b5bed43bda221db22222c3ab3cbee19e8781af72,11 个公开 Markdown 文件(+730/-2)。主要是双语 RFC、R4/R5 及既有成果细化和修订规范的衔接。自修复将 §6 和 E3 补完整:每项受影响租约需要继续、更新依据后继续或停止的处置,缺处置时暂缓受影响变更。原 Turn 仍可恢复尚欠效果一次。
从 128 KiB 声明容量复现出发,M1 验证新增、重启和有界读回。M2 才验一次原子修订与 peer 采用。单独提高 cap 会留下同一存储问题和不可表示的替换历史。
对主干的风险
最危险的实现偏差是旧租约仍提交产物,却被当作满足新要求。当前设计将处置、原结算和成果资格分开验收;独立工作继续,缺依据时扩大核对。快速 premerge 的 4 项直接检查和 4 项自动检查均通过,包含双语、架构及公开边界。环境首跑缺 Node 解析依赖,补齐锁定依赖后重跑通过。按项目配置未查询远端 CI。
我的整体评价
Approval conclusion (author-owned PR; GitHub blocks formal self-approval).
APPROVE。 设计明确了跨轮纠正、在途工作和受影响 peer 的边界。后续的运行时 PR 需要证明实际保存、恢复和采用。
English verdict: APPROVE - The exact documentation head b5bed43bda221db22222c3ab3cbee19e8781af72 defines scoped lease disposition, preserves admitted Turn recovery, and assigns implementation qualification to R4/R5. Four direct and four automated checks passed.
Goal And Delivered Outcome
A long-running Goal can exhaust the shared 128 KiB acceptance declaration budget. Recording a correction only in Vision then leaves canonical acceptance unchanged. The deeper gap is additive accumulation being used for evolving intent: current objective revision, supporting requirements, history and work adoption have different contracts.
This bilingual proposal unifies objective and requirement evolution through the existing Goal authority. It defines source-bound atomic change sets, explicit replacement/withdrawal with retained versions, affected-peer adoption, indexed canonical storage and bounded complete decision context. It connects the existing outcome/alignment RFCs and roadmap R4/R5; no runtime limit, default, installation or active Goal is changed.
1cdd9bbefd1d849ac8aa733bd3d80bc13ad0b722. Submission is based onmainat9790111524bd5fd8b51cc7ac14173707737d5735; the intervening changes do not replace the audited objective/outcome owners.main. No issue is closed by this proposal.The source audit finds existing objective preparation/application/adoption, but no general requirement replacement. Ordinary-context tests deliberately exclude proactive objective-preparation guidance, while replan carries the semantic trigger. This suggests discovery/action-fit gaps; it does not establish how frequently any deployed model invokes the tools. E6 requires natural-language, unprompted tool-choice qualification and subsequent artifact outcomes, including justified no-change and unauthorized-source negatives.
Author Declaration
Self-check: source owners, current roadmap and related open design proposals were inspected; English/Chinese requirements, local links and publication boundary were checked. No model behavior, installed deployment or new storage format is claimed.
Scope And Continuation
Complete within this design scope. M1 delivers an over-capacity Goal correction through real storage, native/CLI and packaged App recovery; M2 delivers coupled objective/requirement changes and two peer adoption/return cycles; M3 qualifies model behavior, shared providers and full costs. Reuse existing R4/R5 and S11 work rather than a competing evolution program. The running capacity failure remains until the implementation and migration qualify.
Bounded refactoring: runtime refactoring is unnecessary for this documentation slice. The design composes existing owners and retains explicit versioned producer/consumer migration instead of introducing a parallel ledger.
Validation
2df49a13ac1e79d295daa2dfd85a39a1060ef1b0.loopx checkwith an isolated empty registry, 11 Markdown files clean; diff whitespaceInitial environment setup exposed missing Node dependencies and ambiguous implicit local runtime roots. Dependencies were prepared in the dedicated checkout; the boundary check then used an isolated explicit registry/runtime. No active runtime was repaired or changed. Baseline characterization is not qualification of the proposal.
Frontend / Visual Evidence
UI impact: none. This changes Markdown design content, not UI/chrome. Packaged desktop/narrow correction, mixed-state and recovery evidence is required by E7 in the implementation slice; no screenshot is presented as shipped behavior.
Type Of Change And Routing
Documentation / architectural proposal. Existing Core Goal acceptance and governed amendment own semantics; R5 owns storage; product/context owners own delivery. Shared-authority runtime conformance and promotion are not changed.