2.0 KiB
2.0 KiB
1. Intent and single-turn executors
- 1.1 Add typed Router input/limits/failure and strict three-enum direct ChatModel implementation with Harness retry auditing.
- 1.2 Add fixed SYSTEM_CHAT executor and protocol-neutral application content/status contracts.
- 1.3 Add KNOWLEDGE_QUERY executor with one lookup call, bounded RAG input, exact reference validation and ID-free public content.
- 1.4 Add focused Router/System/Knowledge tests for mapping, identical retry input, fail-closed routing, one Tool call and fabricated references.
2. PreviousTurn and Run persistence
- 2.1 Add PublishedResultPolicy and centralized limits for bounded write/read projection and RAG document extraction.
- 2.2 Extend DiagnosisRun entity/repository and add V012 migration for intent, release_outcome, published_result and lookup index.
- 2.3 Add ChatRunStore/JPA implementation for prior context, start, intent, safe finish and terminal metrics.
- 2.4 Add focused policy/store/schema tests covering exact safe query, corrupt JSON, fallback exclusion and field leakage.
3. Application use case
- 3.1 Add Diagnosis executor that composes stage 4 Agent and stage 5 release into ID-free typed content and optional PublishedResult.
- 3.2 Implement ChatApplicationUseCase ordering prior reads, Run persistence, observer start/status, routing, dispatch and first-terminal-wins completion.
- 3.3 Add protocol-neutral Run control whose client-disconnect cancellation targets the same RunContext.
- 3.4 Add focused end-to-end use-case tests for all intents, original Query, previous turn, consistent IDs, success/fallback/failure/cancel persistence and observer order.
4. Verification and isolation
- 4.1 Run stage 6A focused tests, relevant stage 2-5 regression tests and Maven compile.
- 4.2 Validate strict OpenSpec, migration/entity alignment, no public Controller/frontend diff, no Router/ordinary executor Tool or Agent loop leakage, and no unsafe PublishedResult fields.