Three Practical Ways to Turn Your Experience into an AI Skill Workflow

Prompt vs. Skill

Prompt engineering solves a one‑off request; a Skill captures a reusable SOP that fixes experience, judgment criteria, and output format so the AI can execute the same process consistently.

1. Provide the Result and Let AI Reverse‑Engineer

When a satisfactory deliverable (plan, article, analysis report) exists but the underlying workflow is undocumented, feed the final result to the AI and ask it to infer a Skill that includes input materials, execution steps, selection criteria, human‑judgment nodes, and output format.

Based on this result, generate a Skill covering inputs, steps, filters, judgment points, and output.

Example – AI Customer Service Knowledge‑Base Optimization plan:

Collect customer service records, FAQs, product docs, and user feedback.

Classify issues: duplicate queries, inconsistent answers, outdated info, unclear processes, unreadable content.

Mark high‑frequency, high‑impact problems.

Identify missing, outdated, or unclear knowledge‑base entries.

Prioritize optimizations by user impact × business value × modification cost.

Output problem list, improvement suggestions, standard answer examples, and an iteration plan.

Future knowledge‑base upgrades can reuse this Skill without re‑analysis.

2. Complete the Task First, Then Ask AI to Summarize

For workflows that are performed repeatedly but remain tacit, execute the task normally while letting the AI record each decision. After completion, ask the AI to distill the captured logic into a Skill.

Example – weekly Xiaohongshu content creation:

Topic: verify concrete user pain, natural product placement, real use cases.

Title: generate 20 candidates, discard overly promotional or vague ones.

Cover: ensure thumbnail clarity, headline hook, color prominence.

Body: start with user problem, expose common mistakes, then present solution.

AI‑summarized Skill:

1 topic.

5 titles.

3 cover directions.

1 article body.

3 comment‑section prompts.

1 pre‑publish checklist.

3. Write a Work Instruction, Then Let AI Optimize

When a mature methodology exists but is not yet engineered, draft a concise work instruction covering five items:

Applicable scenario.

Required materials.

Step‑by‑step execution.

Acceptance criteria.

Desired output format.

Prompt example:

Help me turn this SOP into a Skill, adding boundary conditions, common failure points, and fallback rules.

Example – senior logo designer SOP (client brief, competitor analysis, concept direction, sketch, AI‑assisted generation, refinement, brand‑asset delivery) enriched by AI:

Clients must provide brand name, industry, target users, usage scenarios, and competitor references.

First round focuses on direction only, not final artwork.

Review logo structure before color.

Limit client review to three concepts.

All concepts must be adaptable to business cards, signage, avatars, and packaging.

If a client says "high‑end, atmospheric", AI must ask clarifying questions.

Deliverables must be editable SVG, not raster‑only.

Each concept needs a design rationale.

When merging two concepts, verify visual language consistency.

Universal Skill Template

Skill Name: e.g., "Public Account Article Polishing Skill" 1. Applicable Scenarios – problems this Skill solves. 2. Role Definition – AI role and goal. 3. Input Materials – target, background, assets. 4. Execution Flow – understand goal, organize inputs, decompose task, generate draft, self‑check, output final version. 5. Quality Checks – accuracy, clarity, usability, contextual fit, justification. 6. Output Format – final deliverable, key explanations, optional improvement suggestions. 7. Fallback Rules – handle missing information, unclear goals, conflicting requirements, uncertain facts, or unsatisfactory results.

Where to Store Skill Files

1. Project‑Level Skill (recommended)

Store under the current project so the Skill only affects that project.

YourProject/
├── .codex/
│   └── skills/
│       └── your-skill.md
└── …

2. User‑Level Skill (global)

Place frequently used Skills in a global directory.

# macOS / Linux
~/.codex/skills/

# Windows
%USERPROFILE%\.codex\skills\

Four Skills I Use Regularly

1. Humanizer‑zh

Purpose: Remove AI‑style phrasing. Detects generic filler, exaggerated expressions, and mechanical parallelism. Suitable for public‑account articles, product copy, and personal statements.

Repository:

https://github.com/op7418/Humanizer-zh

2. Content Research Writer

Purpose: Assist deep writing by breaking down article structure, flagging research gaps, data needs, and logical weaknesses. Ideal for long‑form articles, industry analysis, product research, and public‑account essays.

Repository:

https://github.com/ComposioHQ/awesome-codex-skills/tree/master/content-research-writer

3. KKKKhazix (卡兹克写作)

Purpose: Generate long‑form public‑account content while avoiding high‑frequency buzzwords and AI‑style openings. Fits opinion pieces, AI‑related articles, product analyses, and company research.

Repository:

https://github.com/KKKKhazix/khazix-skills

4. Guizang PPT

Purpose: Convert articles into HTML slide decks that can be presented directly in a browser, supporting keyboard and touch navigation. Suitable for sharing drafts, course materials, and pitch decks.

Repository:

https://github.com/op7418/guizang-ppt-skill

Conclusion

AI is often used as a chat box—ask a question, get an answer, then repeat for a new task. The real value lies in extracting the judgment standards accumulated from repeated work and encapsulating them as Skills. Once a task has been performed at least three times, create a rough Skill and iteratively add flow details, edge cases, examples, and failure handling. Skills amplify experience, turning subjective, repeat‑only judgments into stable, AI‑driven processes.