output-mobile-friendly-short
/output-mobile-friendly-short
Dostosowuje odpowiedzi do powierzchni mobilnych: krótkie odpowiedzi czatu, twarde limity słów, bez szerokich tabel i zbędnych zastrzeżeń.
- Autor
- sboghossian ↗
- Repozytorium
- mini-claude-for-legal ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 14.05.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 14.05.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Dostosowuje odpowiedzi do powierzchni mobilnych: krótkie odpowiedzi czatu, twarde limity słów, bez szerokich tabel i zbędnych zastrzeżeń.
Oryginalny opis (EN)
Use when Claude must adapt its output for mobile surfaces — short chat responses, in-app legal answers, or any context where the user is on a small screen or expects a concise result. Enforces hard word limits, removes wide tables, and strips verbose hedging. Triggers on mobile UI signals, short-form question context, or explicit brevity requests.
Jak działa
- When this applies
- Hard limits
- Structure rules
- Confidence signal
- What gets cut on mobile
- What stays — always
- Example — before and after
- Related skills
Podgląd SKILL.md
---
name: output-mobile-friendly-short
description: Use when Claude must adapt its output for mobile surfaces — short chat responses, in-app legal answers, or any context where the user is on a small screen or expects a concise result. Enforces hard word limits, removes wide tables, and strips verbose hedging. Triggers on mobile UI signals, short-form question context, or explicit brevity requests.
license: MIT
metadata:
id: output.mobile-friendly-short
category: output
intent: ["__format__", "mobile", "brevity", "short-form"]
related:
- output-voice-friendly-short
- output-partner-memo-style
- output-irac-structure
priority: P0
source: Louis — HAQQ Legal AI (github.com/sboghossian/mini-claude-for-legal)
version: "1.0"
---
# Mobile-Friendly Short Output
Legal AI on mobile has a fundamentally different contract with the user: they are not reading a memo, they are getting a quick answer — often one-handed, often mid-meeting. Dense paragraphs, wide tables, and exhaustive hedging all fail on a phone screen. This skill governs how Claude compresses without losing accuracy.
## When this applies
Activate when:
- The interface signals a mobile context (narrow viewport, app shell, voice trigger)
- The user asks a quick factual or triage question
- The task is a quick check rather than a formal deliverable
- The user explicitly requests "brief", "quick", "short", or "summarise"
Do **not** activate for:
- Formal document drafting (use [[output-markdown-legal-doc]])
- Full legal opinions (use [[output-irac-structure]] + [[output-partner-memo-style]])
- Research memos with citation requirements
## Hard limits
| Content type | Hard cap |
|---|---|
| Chat / question answers | 300 words |
| Draft documents in mobile | 800 words |
| Tables | Max 3 columns |
| Heading depth | Max H2 (no H3 or deeper) |
| Sentence length | ≤ 15 words preferred |
If an accurate answer cannot fit in 300 words, give the answer in 300 words and offer an expansion: "Need the full analysis? Ask me to expand."
## Structure rules
**Bottom line first.** Every mobile answer opens with the direct answer in one sentence. No preamble, no context restatement.
**Bullets over paragraphs.** Lists are faster to scan on a phone. Use them for any enumeration of 3+ items.
**Cut the context restatement.** Do not echo the user's question back. They know what they asked.
**One alternative, not several.** If options exist, give the best one and note "there are other approaches — tap to see." Do not present 3 parallel drafts.
**Sources at the bottom, not inline.** Citations are important but they disrupt reading on small screens. Collect them: "Sources: UAE FDL 33/2021 art 10; Cabinet Decision 1/2022" — at the bottom, one line.
## Confidence signal
Replace verbose hedging with a single confidence indicator:
| Signal | Meaning |
|---|---|
| ✅ | High confidence — well-established rule, directly applicable |
| ⚠️ | Moderate confidence — rule applies but fact-specific; legal review recommended |
| 🚫 | Low confidence — uncertain, actively contested, or jurisdiction unclear — do not act without qualified legal advice |
Place this indicator at the start of the answer, before the bottom line.
## What gets cut on mobile
- Long introductory context re-statements
- Multiple alternative draft options (offer one + expansion prompt)
- Extended qualification paragraphs ("it depends on X, Y, Z, and many other factors")
- Table of authorities / full citations (compressed to a Sources footer)
- Nested list hierarchies deeper than 2 levels
## What stays — always
- The direct answer
- The key legal rule with its jurisdiction and article reference
- The concrete next step or action
- Any critical warning (jurisdiction mismatch, missing facts, urgent deadline)
## Example — before and after
**Before (desktop-style):**
"This is a nuanced area of law. Under UAE Federal Decree-Law 33/2021 on the Regulation of Labour Relations, specifically Article 43, which deals with notice periods for fixed and indefinite term contracts, an employee working under an indefinite-term contract is entitled to a minimum notice period that depends on the length of service: 30 days for employees with less than 5 years of service, and the period may be extended by the contract..."
**After (mobile):**
✅ UAE notice period (indefinite-term contract):
- Under 5 years service: 30 days minimum notice
- 5+ years: check contract — may be longer
- Source: UAE FDL 33/2021 art 43
Need the full breakdown? Ask me to expand.
## Related skills
- [[output-voice-friendly-short]]
- [[output-partner-memo-style]]
- [[output-irac-structure]]
- [[output-inline-citations-with-pinpoints]]
Wersje
W bazie nie ma innych skilli o nazwie output-mobile-friendly-short.
Migawka bazy z 28.10.2026 · wersja paczki 28b15f128aaa (skrót SHA-256 całej paczki)
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100output-voice-friendly-short
Przygotowuje wyniki pod dostarczanie głosem i czytniki ekranu: krótkie zdania, bez zdań zagnieżdżonych, przypisów, tabel i cytowań w tekście. - 100mobile-friendly-test
Prawdopodobnie sprawdza, czy strona jest przyjazna dla urządzeń mobilnych, na podstawie dostępnej strony albo dostarczonych zrzutów ekranu i HTML. - 90voice-short-spoken-output
Formatuje odpowiedzi do czytania na głos: sam tekst, do 150 słów, krótkie zdania, liczby słownie i rozwinięte skróty.