v1.6.3
This patch release strengthens panda-write for Thai technical, academic, and
professional documents. It adds focused writing guidance and three blind
scenarios for source-grounded Thai output.
Panda Write
- Convert source fragments into complete, natural Thai prose while preserving evidence boundaries and exact technical identifiers.
- State explicit model/table counts before listing exact names; keep model names distinct from database tables.
- Calculate percentages only from confirmed values and avoid unsupported causal, statistical-significance, or performance claims.
- For DOCX documents, use consistent heading styles, omit a TOC unless requested, and indent Thai body paragraphs by one tab stop.
- When a diagram is requested, create and verify Mermaid source before rendering and embedding an image. Keep the source with the document deliverables; do not install a renderer when none is available.
Three blind scenarios cover Thai academic transformations, exact table-count writing, and the DOCX diagram/heading/TOC/indentation workflow. They complement the existing Thai status-note and DOCX template cases.
Verification and limits
On the local 1.6.3 release candidate, the structural validator passed for all 23
skills, local references, manifests, and scenario coverage. The documentation
build created 45 pages; the link checker passed 2,616 internal links. The tooling
suite passed all 30 tests, and git diff --check passed.
These checks validate packaging, documentation, scenario preparation, and repository tooling. Baseline-versus-skill model comparisons for the new Thai writing scenarios are Not Run; their presence does not establish improved generated-document quality. The SQL and writing fixtures are synthetic.
Compatibility and recovery
No application runtime dependencies or migrations are introduced. Plugin manifests advance to 1.6.3. Existing local plugin installations may need an update or reinstall to load the new instructions. The previous source version is available at v1.6.2.