Purpose and target scenarios
This guide is for teams or individuals who want to incorporate AI writing Skills into their editorial process while keeping key decisions human-owned. The referenced Skills map to three broad uses: general production workflows (multi-stage drafting and review), research-led long-form content (source discovery and citation), and WeChat/author-voice workflows (channel adaptation and iterative style learning). Example Skill identifiers from the reference include: contentresearchwriter, ixhlink-skills-article-gen, research-article-crafter, skill-article-evolution, wechat-article-crayon, wewrite, writing-expert-team-pro-sl (listed by purpose, not as installation commands).
Core rules to apply every time
- Define reader and problem: beyond the topic, specify who the piece is for, the problem it answers and the desired reader action.
- Set allowed and disallowed sources: list permitted domains/databases and sources that are not acceptable without review.
- Break work into checkpoints: topic → research/evidence list → outline (approve) → section drafts (review per section) → polish → pre-publish checks.
- Human final sign-off: always verify original sources, asset rights and approve publication manually.
Stage-by-stage inputs and expected outputs (copyable)
1) Topic & audience (input) - Input: one-line topic, reader persona, target length, target channel - Output (acceptance): one-sentence purpose, key reader pain points, expected CTA 2) Research & evidence list (input) - Input: allowed sources, list of key questions - Output (acceptance): for each key claim, provide one or more original-source links; unsupported claims flagged as “to verify” 3) Outline (input) - Input: approved purpose and evidence list - Output (acceptance): section headings, the question each section answers, which evidence items are needed per section 4) Section drafting (input) - Input: approved outline - Output (acceptance): short draft per section (e.g., 300–600 words or channel-appropriate size), claims annotated with reference IDs, clear transitions 5) Editing, polishing and channel adaptation (input) - Input: full draft and channel constraints (WeChat, blog SEO, internal doc template) - Output (acceptance): final draft, metadata (title, summary, cover suggestions), channel adaptation notes 6) Pre-publish check (input) - Input: final draft, all source links, asset attributions - Output (acceptance): fact-check log (numbers, names, dates), copyright/permission confirmation, final approval record
Worked example: “Procurement guide for remote collaboration tools (10–50 people)”
Context: a product manager needs an internal procurement guide recommending suitable communication and document tools for teams of 10–50.
Step-by-step (can be executed in any Agent that supports Skills): 1) Topic brief (example input) - Topic: procurement guide for remote collaboration tools (10–50 people) - Reader: procurement committee and IT lead - Goal: compare options and give 3 prioritized recommendations - Disallowed sources: vendor white papers not in the company procurement DB (require procurement review) 2) Research & evidence (acceptance) - Ask the Skill to list evaluations, vendor docs and customer case studies from the past 12 months, each with original links - Acceptance: each claim such as “Tool X scales to 50 concurrent users” must cite at least one third-party review or customer case; otherwise mark “to verify” 3) Outline (acceptance) - Suggested outline: background & goals → must-have features → budget & deployment risks → tool comparison (3–5) → recommendation & migration steps - Acceptance: each section includes the question it must answer and required evidence items 4) Section drafting & review - Generate each section separately, with in-text reference IDs tied to the evidence list - Human reviewer confirms the accuracy of each data point (prices, concurrency limits, security claims) against original links 5) Polish & channel formatting - Format for internal docs: headings, tables for side-by-side comparison, executive summary (200 words) 6) Publish checks - Manually open each original link to confirm critical numbers and quotes, verify screenshots/graphics permissions, record approver and version
Pre-publish checklist (must run)
- Does every key claim have an original-source link? (yes/no)
- Are numbers, names and dates consistent with original sources? (yes/no)
- Are images/assets authorized for use? (yes/no/needs procurement)
- Is tone appropriate for the channel? (yes/no)
- Are any claims marked as “to verify”? (yes/no)
Limitations and risks
- Skills speed up structured drafting and idea organization but do not replace human verification of sources, licensing, legal or compliance decisions; publication responsibility remains with people.
- Output quality depends on the quality and scope of allowed sources; if permitted sources are biased or incomplete, the output will reflect those gaps.
- Avoid one-shot full-article generation without intermediate acceptance steps — this commonly causes rework and citation errors.
Quick input template to paste into a Skill
- Purpose (1 line)
- Target reader (specific persona)
- Allowed source domains/databases
- Output format (word count, channel, need for summary/table)
- Mandatory acceptance points (list claims that must include original links)
Conclusion Use Skills to automate stages of drafting while enforcing human checkpoints for evidence, tone and rights. Choose a Skill that matches the stage you need (general workflow, research-first, or channel-specific) and run the stage-based acceptance flow to reduce rework and retain editorial control.