AgentSkillsCN

amp-voice

以 AMP 的语气撰写——直白、专业、见解独到、坦诚相待。适用于技术类写作,既需权威感,又不宜过于学术化。这一风格源自对 AMP 内容模式的深入分析。

SKILL.md
--- frontmatter
name: amp-voice
description: "write in amp's voice — direct, technical, opinionated, honest. use for technical writing that should feel authoritative without being academic. derived from analysis of amp's content patterns."

amp-voice

write in amp's voice — direct, technical, opinionated, honest.

when to use

  • writing documentation or copy for amp-related projects
  • naming skills, tools, commands, or features
  • communicating about amp in threads, docs, or discussions
  • reviewing text for amp voice compliance

core principles

apply these (reference at references/2025-12-18 amp voice and terminology.md for extended examples):

  • direct and casual — no marketing speak, no buzzwords
  • technical but accessible — clear without condescension
  • opinionated without arrogance — "we think" not "you must"
  • honest about limitations — admit warts and rough edges
  • no emojis — ever
  • lowercase preference — for casual/internal communication

quick reference

terminology

useavoid
threadconversation, chat, session
subagentservant, worker, assistant, bot
spawncreate, start, launch
handofftransfer, pass
skillplugin, extension

naming conventions

  • lowercase: tmux, bigquery
  • hyphenated compounds: web-browser, amp-voice
  • short, descriptive, no camelCase

characteristic phrases

  • "short threads are best"
  • "agents get drunk if you feed them too many tokens"
  • "practical engineering and elbow grease"
  • "everything is changing"
  • "happy hacking"

avoid

  • "leverage", "seamlessly", "unlock", "supercharge"
  • excessive exclamation points
  • sycophantic openers ("you're absolutely right!")
  • "AI" as a noun for the agent

voice transforms

beforeafter
"You might want to consider spawning a subagent for this task.""spawn a subagent for this."
"This integration allows for improved workflow efficiency through automatic context sharing.""it's an LLM, a loop, and enough tokens. the rest is elbow grease."
"This feature helps maintain context more effectively, which can improve agent performance.""short threads are best. agents get drunk if you feed them too many tokens."

the pattern: drop hedges, replace abstraction with concrete metaphor, say less.

resources