AGENT SKILL

writing-tech-post

Write or edit engineering posts with supported claims. Use for drafts, rewrites, and publication reviews; excludes marketing pages, API references, release notes, and internal specs.

适用职业内容创作者
使用场景写作与文案
License 许可
MIT
所属仓库
compozy/compozy
最近更新
2026-11-13
Stars
★ 2.9k
Forks
188

01 前言 Front Matter

name
writing-tech-post
metadata
{ "author": "Pedro Nauck", "github": "https://github.com/pedronauck", "repository": "https://github.com/pedronauck/skills" }
description
Write or edit engineering posts with supported claims. Use for drafts, rewrites, and publication reviews; excludes marketing pages, API references, release notes, and internal specs.

02 资源文件 25

03 SKILL.md 正文

Writing Engineering Posts

Produce the requested artifact: a new post, an outline, a focused rewrite, or a publication review. Reuse the accepted brief and evidence. Choose a structure that fits the reader and material; a small edit does not restart an entire authoring workflow.

  1. Establish the reader, main point, and supported facts from available context. Ask only for missing information that materially affects the requested output.
  2. For a new post, choose a useful primary archetype and outline. For an existing draft, retain working structure and edit the requested scope. Frontmatter, rung labels, depth tuples, and per-section evidence plans are optional tools unless the publisher requires them.
  3. Explain the mechanism and why it matters. Support quantitative, comparative, security, and capability claims with appropriate evidence and limitations. Distinguish actual measurements, illustrative examples, hypotheses, and work still in progress.
  4. Refine voice and flow where needed. Publisher style matrices, H2 question chains, and first/last-200-word comparisons are editorial heuristics, not universal gates. Do not invent failed mitigations, partial victories, benchmark results, or lessons to satisfy an arc.
  5. Before external publication, apply references/pre-publish-checklist.md to the relevant claims and disclosure obligations. Draft delivery is not publication authorization. The optional linter reports heuristics; human factual review owns unsupported claims and disclosure decisions.

Reference router

Read the reference for the current archetype or editing concern, not every phase:

  • Structure: references/archetypes-and-structure.md; optional depth diagnosis: references/depth-and-abstraction.md.
  • Incident: references/postmortems.md; migration: references/migrations.md; performance: references/performance-deep-dive.md.
  • AI/agents: references/ai-and-agents.md; security/reliability: references/security-and-reliability.md.
  • Evidence/assets: references/evidence-diagrams-code.md; narrative: references/narrative-and-pacing.md.
  • Voice/disclosure: references/voice-and-disclosure.md; a requested publisher register: references/publisher-voice-matrix.md; specific prose problems: references/anti-patterns.md.

Use an assets/outline.<archetype>.md template only when useful; omit irrelevant sections. Run helpers through this skill's absolute directory: python3 <writing-tech-post-dir>/scripts/lint-post.py <draft.md>. --strict makes heuristic warnings fail for an explicitly chosen editorial gate. A clean lint result does not prove factuality or disclosure readiness.

04 同仓库更多 Skills