Writing Minimal Prompts

An explicit-only agent skill for turning notes, briefs and task specifications into the smallest prompt that preserves the intended result.

The skill removes scaffolding that does not change the likely outcome. It keeps scope, authority, decisive evidence, real constraints and stopping conditions when they matter, while leaving method and architecture open unless the user already decided them.

What it prevents

• Roles, sections and workflows added by habit.
• Possible concerns converted into requirements.
• System or repository instructions repeated in the prompt.
• Procedural source material preserved after only the outcome still matters.
• Credentials copied into handoffs.
• Model folklore used without checking current first-party guidance.

Explicit-only by design

This skill does not load for ordinary prompt requests. Invoke it deliberately when you want a prompt shortened, reviewed, written or prepared for handoff:

Use $writing-minimal-prompts to turn these notes into a focused implementation prompt.

Install

npx skills add CodingCossack/writing-minimal-prompts -g -y

For selected agents:

npx skills add CodingCossack/writing-minimal-prompts -g -y \
  -a codex -a claude-code -a gemini-cli -a opencode

Codex native installer:

$skill-installer install https://github.com/CodingCossack/writing-minimal-prompts/tree/main/skills/writing-minimal-prompts

Replaces the model-specific prompt skills

writing-minimal-prompts replaces the retired writing-gpt-5-6-prompts and writing-claude-5-prompts repositories. Model-specific guidance now enters only when it materially affects the task and has been checked against current first-party documentation.

The separate writing-claude-design-prompts skill remains active for prompts whose primary result is a visual artefact, frontend, redesign, prototype, design system or image-to-code implementation.

Structure

skills/writing-minimal-prompts/
├── SKILL.md
└── agents/
    └── openai.yaml

Licence

MIT