skills.rest
wimpysworld
wimpysworldOFFICIAL
Explore all skills by this author →
WebsiteGitHub

write-assistant

Create and update AI assistant and sub-agent system prompts for Claude Code, Codex, Pi, and OpenCode.

716|31|Updated Mar 13, 2023
Search Tags:#system prompt#Prompt Engineering#sub-agent#CODEX#Claude code#Opencode#agent description
One-click install
npx skills add https://github.com/wimpysworld/nix-config --skill write-assistant
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: write-assistant
Source: https://github.com/wimpysworld/nix-config/tree/main/home-manager/_mixins/agentic/assistants/skills/write-assistant
Command: npx skills add https://github.com/wimpysworld/nix-config --skill write-assistant

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Writing effective agent system prompts is hard: prompts suffer from vague instructions, contradictions, wrong voice, bloated token counts, and descriptions that fail to trigger routing. This Skill provides a structured methodology for authoring and maintaining agent prompts that comply with platform conventions.

Core Features & Use Cases

  • Create or update agent prompts: Build a new sub-agent from requirements or surgically edit an existing prompt, with a changelog recording removed, preserved, and added elements.
  • Platform-specific structure: Apply the seven-element template (role, mission, capabilities, process, constraints, output format, examples) with correct frontmatter for Claude Code, OpenAI Codex, Pi, and OpenCode.
  • Voice and trigger guidance: Enforce imperative second-person voice, token budgets, identifier rules, and routing-optimized descriptions with optional <example> blocks.
  • Use Case: You need a new Claude Code sub-agent that reviews TypeScript PRs. Invoke this Skill to generate a complete prompt with a trigger-rich description, constrained tool usage, an output table template, and explicit constraints.

Quick Start

Use the write-assistant skill to create a sub-agent prompt for reviewing TypeScript pull requests with a routing-friendly description and output format table.

Frequently Asked Questions about write-assistant

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I write a system prompt for a Claude Code sub-agent?▼

Use the seven-element template: role, mission, capabilities, process, constraints, output format, and examples. Write in second-person imperative voice, keep it under 700 words, and add 2-4 example blocks inside the description to improve routing.

What should go in an agent description field for routing?▼

Write one sentence that front-loads the use case and includes trigger phrases users might say. For Claude Code sub-agents, add 2-4 example blocks showing context, user message, and trigger reason; keep portable agents as plain prose.

How long should an AI agent system prompt be?▼

Target 400-700 words or 500-3,000 characters, with a hard cap around 1,200 words when examples are included. Terse instructions use roughly 5x fewer tokens and show about 8% better compliance than verbose ones.

Does the same agent prompt format work on Codex and OpenCode?▼

Codex prefers a single instructions string in YAML, while Pi and OpenCode use the portable two-field frontmatter of name and description. Example blocks degrade gracefully on non-Claude platforms but add token noise.

When should I include examples in an agent prompt?▼

Include 1-2 compact examples when output style is subjective, judgment decides inclusion, or the format is too complex for prose. Omit examples for procedural tasks with one correct output form or schema-constrained output.

Why does my agent prompt cause over-triggering or poor compliance?▼

Aggressive caps like IMPORTANT or YOU MUST cause over-triggering on Opus 4.5+ and Sonnet 4.6. Replace them with plain imperatives, remove first-person phrasing, and resolve contradictions such as be concise alongside be thorough.

Related Skills

Explore More Software Engineering Skills→
treasure-datatreasure-data
21

agent-prompt

Generate structured system prompts for Treasure Data AI agents.

Official
Basic
dioerdendioerden

coding-agent

Execute coding tasks with Codex, Claude Code, OpenCode, and Pi agents.

Community
Advanced
githubgithub
38.5k

create-agentsmd

Generates an AGENTS.md file with setup, testing, and workflow instructions for repositories.

Official
Basic
RafaellsAlmeidaRafaellsAlmeida

bmad-agent-cis-storyteller

Enforce a storyteller agent persona during interactive narrative sessions.

Community
Intermediate