Repository navigation
Dazzler 0.30.0 — Offline Design Checks and Reliable Exports
Dazzler 0.30.0 adds offline design-evidence tools, hardens document exports and refreshes all thirty examples.
Changed
- Updated the pinned Style Science consumer and added an optional offline wrapper for capabilities, advisory evidence and measured comparisons. It does not call an AI provider or authorize automatic changes.
- Rejected unsafe archive aliases, Windows device names and file/parent collisions before restore writes begin.
- Fixed single-row PowerPoint table loss, malformed table handling and heading-case audits across HTML, Word and PowerPoint line breaks.
- Rebuilt all ten Word, ten HTML and ten UI examples, with updated field-guide and README snapshots, themeable sources and preserved fictional data.
Added
Regression checks for hostile paths, empty and paginated tables, native heading boundaries, stale or missing model evidence, empty searches and repeated order clearing. Every gallery example has current render and comparison evidence.
Why
Prevent silent export errors, keep design decisions grounded in available evidence and demonstrate the current skill with a varied gallery.
Verified
- All six hosted CI jobs passed, including Windows/Linux, Node 20/22 and byte-identical cross-platform builds.
- 105 Node tests passed; 91 Python tests completed with one optional skip.
- Ten native documents rendered as ten pages. Twenty browser examples passed heading, axe, overflow and applicable interaction checks at desktop and mobile widths.
- All 121 bundled browser font files loaded; field-guide/gallery, integrity, pinned-source, formatting and link checks passed.
- All ten archives passed extraction and helper checks. The largest host package is 12.9 MB compressed / 22.4 MB unpacked, below the 24 MB gates.
- The Windows x64 offline kit restored into a clean directory and rebuilt all three bundles without installation scripts or network downloads; its 1,709-file integrity check passed.
Installation
Use the ZIP for your host with install_skill.py and SHA256SUMS.txt. The offline build ZIP and manifest.json are maintainer downloads, not skill-upload packages. Your agent's ordinary brand constraints and rendering checks remain authoritative.
Notes
Visual review is agent evidence, not a beauty score. Native pagination depends on the host and later edits. Optional R integration and authenticated cloud-upload acceptance remain separate checks. Font binary metadata reinspection was unavailable locally; unchanged binaries passed inventory and browser-load checks. Original licenses and attribution remain included.