Stats
Actions
Tags
From st
Use when executing any command or agent task. Cross-cutting principles for all Superteam work. Auto-triggered by all commands and agents. Covers: visual-first (design decisions, verification, comparisons, any visual content), UI screenshots, Playwright, browser preview, HTML preview, questioning, one question per message, adaptive, ASK/PRESENT/CONFIRM, recommendation, decision continuity, confirmed decisions, DECISIONS block, never re-ask confirmed.
How this skill is triggered — by the user, by Claude, or both
Slash command
/st:core-principlesThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Cross-cutting principles for ALL commands and agents. Load progressively based on context.
Cross-cutting principles for ALL commands and agents. Load progressively based on context.
Core Principles > Research findings > Agent preferences.
Research informs WHAT to propose, never HOW to present. Rules in this file and its references are NEVER overridden by research context. "Technically compliant" loopholes = violation. Rules apply to spirit, not just letter.
See: references/research-boundaries.md for concrete examples, anti-rationalization table, and decision point reminder.
When Principle 1 (Visual-First) and Principle 2 (Questioning) BOTH apply: → Principle 1 fires FIRST on FORMAT. → Create visual content FIRST, then ask using Principle 2.
Example: Presenting design options = visual + questioning. Create preview of ALL options FIRST, then show screenshot and ask "Which do you prefer?"
When it can be shown, show it.
Text tells you what SHOULD look like. A preview tells you what ACTUALLY looks like.
When it applies: Design choices, implementation results, comparisons, diagrams, any visual content better understood visually than as text.
Core idea: When content is better understood visually than as text, create a visual representation BEFORE asking for decisions.
Quick summary: Can be shown? → Show it. Before deciding AND after building. Design dimensions are VISUAL — preview PER DIMENSION inline when proposing. Light background default.
Action (always, before proposing any visual choice): Create HTML preview at .superteam/preview/<name>.html showing ALL options as side-by-side visual cards → serve with python3 -m http.server → take screenshot → present screenshot → THEN ask user to choose. Text-only proposals for visual choices = violation.
See: references/visual-first.md for full principle, execution strategies, and anti-patterns.
When it applies: Every interaction with the user. Guiding exploration, narrowing scope, making decisions.
Core idea: Focused, adaptive questioning based on facts. Recommend with reasoning.
Quick summary: One primary question per message (closely coupled questions OK, max 2). ASK/PRESENT/CONFIRM interaction types. Exploration phase: open-ended, no options. Narrowing phase: fact-based multiple choice. Never predetermined scripts.
See: references/questioning.md for full principle, interaction types, and adaptive flow.
When it applies: After any user confirmation. Within a conversation and across subagent handoffs.
Core idea: Decisions are permanent. Use file-based persistence (.superteam/decisions.json) so decisions survive context truncation and transfer across agents.
Quick summary: Before proposing ANY value that might have been decided → read .superteam/decisions.json. If found, use it and state "Using [value] (confirmed earlier)". After confirmation → write to decisions.json immediately. Maintain in-context DECISIONS block as cache.
See: references/decision-continuity.md for full principle, file schema, and write/read patterns.
| File | ~Lines | When to Load | Trigger |
|---|---|---|---|
SKILL.md | ~97 | Always | Auto: all commands and agents |
references/research-boundaries.md | ~45 | When reading research files or making decisions from research | Commands with research input |
references/visual-first.md | ~165 | Design choices, previews, implementation verification | Present/CONFIRM interaction involving visuals |
references/questioning.md | ~261 | All user interactions | ASK/PRESENT/CONFIRM |
references/decision-continuity.md | ~258 | Before proposing values, after confirming | Read decisions.json, update decisions.json |
Progressive loading: SKILL.md gives principle summaries + cross-principle priority. Load reference files only when working on specific principles. Most tasks need SKILL.md + 1-2 reference files.
- Follow `superteam:core-principles` for all work.
- Load principle references on-demand via context trigger.
Used by: All commands and agents.
Pairs with:
superteam:verification — visual-first for UI checkssuperteam:requesting-code-review — visual-first when reviewing UI changessuperteam:project-awareness — detection context informs questioningsuperteam:research-methodology — research area selection uses questioning for approvalsuperteam:handoff-protocol — decisions.json transfers confirmed decisions across handoffsnpx claudepluginhub tuanldas/superteam --plugin stDefines autonomy boundaries for AI-human teams: users decide product vision, priorities, and scope; AI experts autonomously handle technical architecture, design, frontend/backend implementation without asking questions. Use on every task.
Crafts actionable design principles for teams to guide decisions and resolve trade-offs, with qualities, structure, 6-step process, and best practices.
Entry point for the Designpowers design workflow system. Manages welcome sequence, taste profile check, and routes to the correct first skill. Activates on design-related messages.