- Full task-by-task comparison of current code vs roadmap requirements
- notebook-navigator cross-reference completion status
- Architecture advancement summary since v1.8.2
- Current module map and verification gate
- Hard constraints status (still active)
- Next direction: command consolidation + runtime packaging
- CI status: all clean, no errors to fix
- Bilingual: EN + ZH-CN progress audit documents
- Roadmap docs updated with 2026-05-02 progress
Task 2 (promptUtils.ts legacy prompt): Any extension/retirement MUST
fully preserve original scenario usability. Cross-version stability
and UX continuity take priority over cleanup.
Task 3 (MermaidProcessor decomposition): Each sub-task MUST be
individually verified in real Obsidian before proceeding. Output images
MUST be saved, checked, confirmed complete. Unit tests alone are not
sufficient to advance beyond any sub-task boundary.
Updated: audit docs (EN+ZH-CN), roadmap docs (EN+ZH-CN).
resolveProviderTokenLimit now matches Cline behavior for unknown models:
when global maxTokens equals default (8192) and model is not in the
known metadata table, return undefined — letting the API provider decide.
User-custom maxTokens for unknown models remain preserved (backward compat).
normalizeSpec now handles multiple edge field conventions:
source/target, sourceId/targetId, start/end → from/to.
buildDiagramSpecPrompt explicitly instructs LLMs about field names.
Adds progress audit brainstorm docs (bilingual) covering:
- LLM backward compatibility matrix across all 5 transports
- Current architecture status vs roadmap plan requirements
- Post-v1.8.2 direction and scope boundaries
Updates both roadmap docs with 2026-05-01 hardening section.
All CI-equivalent checks pass.
111 suites, 594 tests. Build, audit:i18n-ui, audit:render-host, git diff --check.
