8094+真实DB验收通过:五环①②③④(含引用列 OWNER→SYS_USER)+ AI ①面(READY_FOR_HUMAN_REVIEW)
+ 部署包(requiresHumanApproval) + 红线全守;实证写入脚本头注。
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Brainstorm output. Locks: NocoBase-style modeler+live-designer (no linear wizard);
v1 builds NEW features only (CREATE proven, ALTER=M4 deferred); recipe auto-generates
wired default page then edit in DesignerPage; AI-assist via external agent+Skills
(no in-platform LLM); single metadata truth SYS_ENTITY/SYS_ENTITYATTRIBUTE.
Companion to 2026-06-19 direction+roadmap. Ready for writing-plans.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
sys_page_schema standard layer had two runtime writers: the converter import CLI and AiLocalReleaseExecutor's raw JDBC INSERT (the critic-flagged net-new bypass). Introduce PageSchemaWriter as the single runtime write function (typed-recipe validation warn-not-block + etag + UPSERT), and route both a new admin-gated PUT /api/page/schema/{id}/standard endpoint (net-new producer for the design-center 'new feature' path) and AiLocalReleaseExecutor.applyPageSchema through it. No remaining raw INSERT into sys_page_schema in the AI runtime write path.
Scope: page_schema runtime write only. The deployment-package SQL-text generator (AiAgentService.pageSchemaSql, offline human-apply artifact) and SchemaImportCli (batch converter import) are separate legitimate paths. AiLocalReleaseExecutor sys_serviceflow/sys_menu raw writes remain (follow-on). Completion = code-complete, integration-pending (end-to-end render proof env-gated).
Tests: PageSchemaWriterTest (4) + PageSchemaControllerTest putStandard success/403/split (27 total) + AiLocalReleaseExecutorTest/AiAgentApiPipelineTest unchanged-green. TDD.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Commit the architecture-direction design doc (D1-D7 locked decisions) and add a companion living execution roadmap reconciling the design with the already-merged AI agent module (main 9aea6c1): phase-by-phase delivery account + ordered milestones (M0 component L1 -> render-smoke gate -> 先宽度 M1/M3a/M2/M3b -> 再深度 M4/M5/M7/M8), split autonomous-build vs env-verification.
Roadmap revised after adversarial review: restored the D2-locked P0 prerequisite (component L1), moved the render-smoke correctness gate ahead of real-DB dogfood, split net-new write-path convergence ahead of the new endpoint, marked code-complete vs integration-pending.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Per the architecture-review redline (EIP3 决策会纪要: AI claims limited to 'AI-assisted, human-reviewed'; never promise 'AI full-auto development'), soften endpoint @Operation summaries, the default task objective, and AI docs that read as committed full-automation. Endpoint paths/method names unchanged (API contract). Added a capability-boundary disclaimer at the top of AI_AGENT_AUTO_DEVELOPMENT.md. 33 platform-ai tests green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
AiAgentController's endpoints relied only on the SecurityFilterChain (prod: /api/** authenticated; dev: permitAll), so any authenticated user could trigger AI writes/actions (auto-develop, local-apply, change-sets, deployment packages) - a lower bar than the comparable overlay/standard-page write endpoints which require admin.
Gate the 14 mutating/action POST endpoints with requireAdmin() (UserSession CURRENTUSER == ADMIN), aligned with PageSchemaController.requireAdmin / AdminWriteController. Read GETs and the pure validate-schema/access-check endpoints stay open. Actual three-truth writes are now admin + localhost double-gated.
Tests: AiAgentControllerTest + AiAgentApiPipelineTest run under an ADMIN session (+ AfterEach clear); added a 403 test for non-admin. 61 platform-ai tests green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
putOverlay hard-rejected (400) any overlay using components/actions outside the ~35-component AI recipe whitelist. Existing converted pages use 13 unregistered component types (search-box/dialog/mapping/tabs/each/tree-select/input-date-range/…) plus 6 onEvent action types that are structurally actionType-without-type (ACTION_WITHOUT_COMPONENT), so editing+saving virtually any existing list page in the designer returned 400.
Split path: putOverlay (designer write path) now validates-and-warns instead of blocking; the strict typed gate stays on POST /api/page/schema/validate and the AI draft pipeline. Clears the only existing-page regression blocking the ai-agent merge to main. TDD: PageSchemaControllerTest#20 rewritten RED->GREEN; 37 schema tests green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>