Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Open Orbit briefing skill — selected by the Orbit pipeline when Gmail is the user's only connected connector, or when the user explicitly scopes their daily digest to Gmail. Pulls the past 24 hours of inbox activity (replies awaited, mentions, cc, auto- categorized bulk) from the user's authenticated Gmail connection and renders the digest as the Orbit Daily Digest email opened inside Gmail's reading view. This skill should not be triggered manually — it is invoked by Orbit's daily-digest schedu
.claude/skills/nexu-io-orbit-gmail/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 21% | 0% |
| case-23 | ✗→✓ | ▲ Improved | 61% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 117% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 25% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 531% | 0% |
Single-connector Orbit template scoped to Gmail. The briefing renders as the Orbit Daily Digest email already opened inside Gmail's reading view — Gmail top header + the email chrome (toolbar / subject / sender / digest body / reply bar). There is no left rail, no inbox list, and no three-pane layout.
Step 1. Open and read the shipped example.html in this folder before writing any output. That file is the canonical design — your job is to reproduce it, not reinterpret it.
Step 2. Mirror the example's structure 1:1:
<header> (Gmail top bar) →<main class="digest-wrap"> → <div class="email-chrome"> → toolbar / subject / sender row / digest body / reply bar.
(hamburger / wordmark / search bar / help / settings / app launcher / avatar). Do not add a left rail (no Compose button, no system labels, no Categories tabs, no colored label list).
digest email is shown.
strip → 需要处理 → 值得关注 → 仅供知悉 → digest footer.
<script> block at the end (action-btn / reply-btn linkinjection).
Step 3. You may refresh mock copy (sender names, subjects, summary text, times) so it reads as "today", but you must not invent extra UI: no inbox listing, no left rail, no Categories tab strip, no extra digest sections, no chrome ornaments. If a detail is not already in example.html, it does not belong in your output.
The sections below are a reference for tokens and visual language — not a license to extend the page.
This skill ships with its own complete visual language baked into example.html (Gmail / Google Sans / Material chrome). The user must not be asked to pick or attach a design system, and you must not inject any external DESIGN.md tokens into the output.
example.html.This is a hard constraint: the briefing must read as a real Gmail page, not as the user's brand.
page bg: #f6f8fc
surface: #ffffff
border: #e0e0e0
text: #202124
text-secondary: #5f6368
text-muted: #80868b
surface-hover: #f1f3f4
red (Gmail): #D93025 /* Compose, important markers, accent */
blue: #1a73e8 /* CTA / link */
yellow: #f4b400 /* important ★ */
green: #0f9d58
search bar bg: #eaf1fb /* light blue-tinted pill */Type stack:
'Google Sans', 'Roboto', -apple-system, system-ui, sans-serif<header>) — full width, white.Left: hamburger (☰) + Gmail wordmark (Gmail, first G red). Center: rounded search bar (#eaf1fb bg, search icon left, settings icon right, placeholder 搜索邮件). Right: ❓ help, ⚙ settings, ▦ Google apps launcher, round avatar.
<main class="digest-wrap"> <div class="email-chrome">)— the opened email lives directly under the header. No left rail, no inbox list. Sub-blocks in order:
a. Email toolbar — back / archive / delete / mark unread / label / spacer / prev / next.
b. Email subject area — <h1 class="email-subject"> with the digest subject (e.g. ☀ Eli, 你昨天的 6 封重要邮件 — Open Orbit Daily) followed by an inline Orbit tag.
c. Sender row — round avatar O + Open Orbit <orbit@opendesign.local> + 收件人 我 ▾ + date right-aligned + reply icon + more icon.
d. Digest body (<div class="digest-body">):
action-btn primaryaction-btn ghostdigest-footer micro-tage. Reply bar — bottom row with 回复 / 全部回复 / 转发 buttons.
where they appear in the example.
belongs to the Orbit digest only.
| Don't | Do | |---|---| | Render a left rail (Compose / system labels / colored labels) | Skip the rail entirely; the page is single-column under the header | | Render an inbox list of other emails | Show only the opened Orbit Daily Digest email | | Render a Categories tab strip (主要 / 社交 / 推广) | Skip it; the digest occupies the reading view directly | | Use non-Google typography | Use 'Google Sans', 'Roboto', -apple-system, system-ui, sans-serif | | Add drop shadows on the Gmail chrome | Flat surfaces; only the subtle Material 1 elevation when an element is focused | | Render avatars as squares | Always circles — sender 40px, card 32px, inline 28px | | Use lorem ipsum | Write real-shaped Gmail copy: "Q3 预算确认", "Login redesign 反馈", senders like Allen Liu / Marie / Nina Park | | Use dark mode | Stay on Gmail's default light theme (#f6f8fc page) | | Brand the Gmail chrome with Orbit | Orbit branding lives only inside the digest body (subject Orbit tag + footer micro-tag) | | Put yellow important stars on multiple inbox rows | Only the Orbit Daily Digest row can carry the important marker |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-03 | fail→fail | 51,724 | 41,025 | -21% | 1 | 1 | 0% | 8,233 | 9,936 | +21% | 0 | 0 | — |
case-01 | fail→fail | 50,043 | 60,855 | +22% | 1 | 1 | 0% | 8,246 | 9,949 | +21% | 0 | 0 | — |
case-02 | fail→pass | 50,705 | 49,577 | -2% | 1 | 1 | 0% | 8,242 | 9,945 | +21% | 0 | 0 | — |
case-04 | fail→fail | 51,758 | 31,619 | -39% | 1 | 1 | 0% | 8,252 | 9,955 | +21% | 0 | 0 | — |
case-05 | pass→fail | 24,324 | 38,257 | +57% | 1 | 1 | 0% | 3,758 | 9,956 | +165% | 0 | 0 | — |
case-06 | pass→fail | 75,728 | 52,418 | -31% | 1 | 1 | 0% | 8,253 | 9,956 | +21% | 0 | 0 | — |
case-07 | pass→fail | 11,596 | 51,230 | +342% | 1 | 1 | 0% | 2,039 | 9,914 | +386% | 0 | 0 | — |
case-08 | fail→fail | 33,252 | 35,928 | +8% | 1 | 1 | 0% | 7,279 | 9,943 | +37% | 0 | 0 | — |
case-23 | fail→pass | 37,689 | 50,746 | +35% | 1 | 1 | 0% | 6,181 | 9,937 | +61% | 0 | 0 | — |
case-09 | fail→pass | 19,940 | 37,980 | +90% | 1 | 1 | 0% | 4,578 | 9,954 | +117% | 0 | 0 | — |
case-10 | fail→fail | 20,734 | 47,953 | +131% | 1 | 1 | 0% | 4,355 | 9,977 | +129% | 0 | 0 | — |
case-11 | fail→pass | 23,987 | 22,804 | -5% | 1 | 1 | 0% | 5,308 | 6,640 | +25% | 0 | 0 | — |
case-12 | fail→fail | 12,554 | 26,816 | +114% | 1 | 1 | 0% | 1,979 | 6,147 | +211% | 0 | 0 | — |
case-13 | fail→fail | 19,379 | 11,217 | -42% | 1 | 1 | 0% | 3,247 | 4,045 | +25% | 0 | 0 | — |
case-14 | fail→fail | 21,841 | 35,709 | +63% | 1 | 1 | 0% | 3,462 | 9,949 | +187% | 0 | 0 | — |
case-15 | fail→pass | 11,346 | 38,399 | +238% | 1 | 1 | 0% | 1,576 | 9,946 | +531% | 0 | 0 | — |
case-16 | fail→fail | 16,713 | 51,262 | +207% | 1 | 1 | 0% | 3,460 | 9,941 | +187% | 0 | 0 | — |
case-17 | fail→fail | 49,173 | 83,027 | +69% | 1 | 1 | 0% | 8,248 | 9,951 | +21% | 0 | 0 | — |
case-18 | fail→fail | 21,213 | 22,507 | +6% | 1 | 1 | 0% | 3,264 | 6,984 | +114% | 0 | 0 | — |
case-19 | fail→fail | 10,778 | 20,549 | +91% | 1 | 1 | 0% | 1,862 | 5,790 | +211% | 0 | 0 | — |
case-20 | fail→fail | 12,669 | 49,790 | +293% | 1 | 1 | 0% | 2,890 | 9,959 | +245% | 0 | 0 | — |
case-21 | fail→fail | 41,074 | 30,558 | -26% | 1 | 1 | 0% | 8,242 | 8,363 | +1% | 0 | 0 | — |
case-22 | fail→fail | 30,794 | 35,051 | +14% | 1 | 1 | 0% | 6,955 | 9,944 | +43% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +9 percentage points is the difference between those two pass rates over the 23 comparable cases. 5 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.