Skip to content
Back to skills

Brainstorming

ASecurity

Use when starting a new project or large feature, or when a change needs unresolved design decisions. Skip read-only tasks and straightforward changes with settled requirements.

  • 2 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added October 2, 2026
ai-agentsgorefactoringsecurity

Security analysis

A100/100

Pro scans all 9 files and shows the line behind each finding

Scanned October 2, 2026

npx -y skills add lsy041015/orchestra --skill brainstorming --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Brainstorming?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Brainstorming
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/lsy041015-brainstorming/badge)](https://www.skillsdirectory.com/skills/lsy041015-brainstorming)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: brainstorming
description: Use when starting a new project or large feature, or when a change needs unresolved design decisions. Skip read-only tasks and straightforward changes with settled requirements.
---

# Design before implementation

The main session owns design. Preserve the user's purpose, constraints and acceptance conditions. Read the relevant existing flow before choosing an approach; do not repeat questions already answered in the request.

**Kickoff.** When starting a new project, or a large feature (more than three files, or a design the codebase does not have yet) with no spec or plan yet, do not infer the goal. Ask until the deliverable, success criteria, constraints, non-goals and target user or environment are clear (skip what the request already answers), summarize them back, and get the user's approval before design or planning. This is the one intake gate; its approval covers the plan that follows. Batch the questions; after two rounds, or when the user says to proceed, record the open assumptions in the summary and continue. A dispatched worker skips this gate.

Choose only the depth the task needs:

- **Bounded change:** requirements and implementation boundary are clear, and the change is within the kickoff's size line (three files or fewer, no design the codebase lacks). State the approach briefly when useful, then implement within the existing authorization. No mandatory spec file, plan file or separate approval turn.
- **Feasibility spike:** identify the concrete question and cheapest adequate probe. Read-only checks can proceed. Keep throwaway experiments isolated and label their limits; retaining prototype code requires production verification.
- **Architectural change:** new subsystem, cross-component interface, migration or meaningful unresolved tradeoff. Read [architecture decisions](references/architecture.md), record the design and acceptance conditions, and use `orchestra:writing-plans` when a multi-step implementation plan helps execution. A large feature that passed the kickoff always gets a plan this way; do not implement it inline from the kickoff summary.

After kickoff, ask only for missing information or a consequential decision that cannot reasonably be inferred. Existing authorization persists. Honor an explicit request to stop for design or plan approval; otherwise do not turn routine document creation into a new permission gate. Host rules govern destructive or external actions. Record newly discovered scope or risk and reassess only the affected decision.

Stay within the request. Preserve security, privacy, data integrity, accessibility, compatibility and hardware calibration. Avoid unrelated refactoring and speculative features. Use a visual only when it clarifies a real decision, with the available host tools; do not start a separate visual-companion server by default. When the user asks for the browser companion, follow the [visual companion guide](visual-companion.md).

Before implementation, confirm every acceptance condition has a verification path. For behavior changes use meaningful regression/TDD checks; use `orchestra:verification-before-completion` for reusable evidence. No separate planner or reviewer agents.

Files in this skill

  • SKILL.md3.2 KB
  • agents/openai.yaml114 B
  • references/architecture.md1.4 KB
  • scripts/frame-template.html7.7 KB
  • scripts/helper.js5.2 KB
  • scripts/server.cjs23 KB
  • scripts/start-server.sh7.7 KB
  • scripts/stop-server.sh4 KB
  • visual-companion.md13.1 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…