一键导入
write-docs
Write documentation in the Beyond voice. Use when writing API docs, guides, tutorials, READMEs, or reference material.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Write documentation in the Beyond voice. Use when writing API docs, guides, tutorials, READMEs, or reference material.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Write or improve ARCHITECTURE.md files for packages and systems. Use when creating documentation for a new package, reviewing existing architecture docs, or when asked to document how a system works.
Audit production readiness of a codebase area. Evaluates tracing, metrics, structured logging, tests, benchmarks, error handling, health checks, graceful shutdown, and configuration. Findings scored with ICE model. Use when asked about production readiness or operational maturity.
Security audit focused on authentication, session management, JWT/token handling, SQL injection prevention, password storage, and Rust-specific CVE patterns. Fetches OWASP reference material before auditing. Findings scored with ICE model. Use when asked to review security posture, audit auth code, or evaluate a codebase for common vulnerabilities.
Deep technical audit of a codebase area. Auto-detects language persona (rustacean/gopher/product engineer), evaluates architecture, operations, performance, security, and idioms. Findings scored with ICE model. Use when asked to audit, review quality of, or evaluate a package or directory.
Audit test coverage, quality, and gaps for a codebase area. Identifies missing critical tests, evaluates test patterns, and scores gaps with ICE model. Use when asked about test coverage, missing tests, or testing quality.
| name | write-docs |
| description | Write documentation in the Beyond voice. Use when writing API docs, guides, tutorials, READMEs, or reference material. |
| allowed-tools | Read, Glob, Grep, Edit, Write |
| model | claude-sonnet-4-6 |
Load the verbal identity:
Read: .claude/skills/brand/verbal-identity.md
Your job is to teach the reader how to do their job.
Not to explain the system. Not to describe your architecture. Not to prove you thought about edge cases. The reader arrived with a goal. Get them there.
Every sentence earns its place by answering: does this help the reader accomplish their goal faster? If not, cut it.
Different doc types serve different jobs. Mixing them produces mush.
The README is a discovery document. The reader just found this package and is deciding whether to go deeper. They have one job: evaluate fit.
Structure:
Rules:
Example opening (correct):
# beyond/auth
Authenticate users, issue tokens, and manage sessions — deployed inside your network, owned by you.
## Quick Start
```sh
cargo add beyond-auth
```
let session = client.sessions().create(user_id).await?;
Example opening (wrong):
# beyond/auth
beyond/auth is a comprehensive authentication library that provides a full suite of identity management features including session handling, token issuance, and user authentication built on modern cryptographic primitives...
Guides are task documents. The reader has a specific job to do and needs a path through it. They don't want an overview — they want the steps.
Structure:
Rules:
session_id in the response")Reference is a lookup document. The reader knows what they want to do — they just need the exact parameter name, the valid values, the shape of the response.
Structure:
Rules:
Neutral, precise, technical. Assume the reader is competent.
| Aspect | Approach |
|---|---|
| Tone | Neutral, direct |
| Assumptions | Reader is a developer |
| Length | Minimum viable |
| Code:Prose | More code, less prose |
| Warmth | None needed |
Don't explain what developers already know. Don't define HTTP. Don't say "simply" or "just" or "easy." Don't hedge with "you may want to" when you mean "do this."