v0.6.0 — token-lean emission, same beautiful reports

Same beautiful reports, meaningfully fewer emitted tokens — plus the business visual system.

Token economy

Measured on the bundled examples, per-report emission drops 12–20% from the compact style alone and ~25–30% with the new contract features; SKILL.md context cost drops 35%.

• aio:table@1:csv — a CSV body profile for pure data tables (~20–30% fewer bytes than minified JSON). Info strings now accept aio:<name>@<major>[.<minor>][:<body-format>].
• waterfall@1 — kind is derived from position/sign; subtotal/end values are derived from the running cumulative (labels and CSV export show the derived numbers, eliminating agent-side arithmetic drift).
• chart@1 — series[].name optional, falling back to the chart title or a locale default.
• callout@1 — body optional when items is non-empty.
• SKILL.md rebuilt for compact emission (18.9 KB → 12.2 KB): compact few-shot examples agents imitate, explicit emission-economy rules (compact JSON, no figure restatement in prose, never hand-write an executive summary — the runtime auto-builds one, omit every derivable field), and a candidate-component key-reference table.

All previously valid documents validate and render unchanged — every contract change is required→optional or additive.

Business visual system

Gray-blue business palette, navy report header, white fine-border cards; the shell grows to 1800 px on wide screens and every component reflows on phones (plain table@1 rows become labeled field cards; 44 px tap targets). Strict Light/Dark two-state theme. New --expire-hours <N> render flag injects a live countdown banner for time-limited download links.

Install

<script src="https://cdn.jsdelivr.net/gh/wxkingstar/ai-output-runtime@v0.6.0/assets/ai-output-runtime.js"></script>

npx skills add wxkingstar/ai-output-runtime

Full details in CHANGELOG.md. Live demo: https://wxkingstar.github.io/ai-output-runtime/

🤖 Generated with Claude Code