Repository navigation
Satori v6.6.0: Landing Page Overhaul, Modern Web Chat Skills Ecosystem & Pipeline Hardening
Versioned as v6.6.0. Major landing page redesign, modern Web Chat Skills ecosystem integration, live frontier judge calibration against Gemini 4 Argon, and full release engineering hardening.
1. Landing Page & Web Presence Overhaul (docs/index.html & docs/404.html)
- WCAG 2.2 AA Adaptive Theming: Implemented light/dark theme CSS variables, auto-detecting system preference with an accessible toggle button and inline SVGs.
- Editorial Hero Layout: Refreshed hero with canonical "Built to be outgrown." callout, dual-track install card (Full Zip top billing,
npx skills addcopy bar secondary, single anchor jump to#install), 4-column metric strip, and vector hero banner. - Surface Chooser & Worked Example 1: Reordered surface chooser cards and refactored Worked Example 1 into a conversational card with speaker avatars and dialogue bubbles.
- Modern Web Chat Skills Cards: Restructured installation into 5 subsections featuring visual setup cards for Claude.ai, Google Gemini Web, and ChatGPT with official November 2026 support documentation links.
- Streamlined Proof & Responsive Tables: Retired 4 legacy benchmark cards in favor of a direct link to repo
evals/and live Tessl endpoint badge; wrapped Token Budgets and Verification suites in accessible<details>accordions; enforcedmin-width: 620pxtable scroll floor. - Standalone 404 Page: Created self-contained, responsive
docs/404.htmlwith zero em dashes.
2. Crisis Safety Routing Parity
- Replaced dead IASP directory URL with canonical
https://www.iasp.info/crisis-centres-helplines/across all 15 occurrences in 13 files.
3. Release Pipeline & Manifest Parity
- Updated
scripts/build_release.pyto track and synchronize all 9 ecosystem manifests, strictly enforcing 255-character descriptions andplugins[0]["version"]. - Rebuilt distribution archives in
dist/and generated updateddist/SHA256SUMS.txt. - Hardened static validator
scripts/validate_skill.pywith automated manifest parity assertions (passing 58/58 checks).
4. Live Frontier Judge Calibration (Gemini 4 Argon)
- Evaluated 50 human-labeled response pairs debiased against Gemini 4 Argon: achieved Cohen's
$\kappa = 0.96$ (98% observed agreement rate) surpassing the$\ge 0.80$ release target.
5. CI Automation & Hygiene
- Added automated external link validator
scripts/check_external_links.py(65/65 passed) and weekly GitHub Actions workflow. - Added fast-forward tag immutability check to
.github/workflows/publish.yml. - Closed out Decision D-2 waiver in
CHANGELOG.md. - Pruned 14 merged stale remote branches.
Verification
- Static Gates: 58/58 PASSED
- Memory Engine: 8/8 PASSED
- Trigger Evals: 98.0% non-activation rate, 100% crisis safety
- External Links: 65/65 PASSED