Turn an SOP into a reusable AI skill in 25 minutes

Steps

1. Package the process

   Open your agent in the empty folder and attach the SOP, just as you would supply samples for a writing-style skill.

   Copy, then replace the [bracketed] partsCopy prompt

   Read [SOP file]. Create process-skill/SKILL.md with YAML frontmatter containing name set to process-skill and a description explaining what it does and when to use it. Copy the approved SOP into process-skill/references/sop.md and reference it from SKILL.md. Preserve the source version and date. Write ordered instructions covering required inputs, decisions, exceptions, output format and acceptance checks. Preserve human approval gates. Flag contradictions and missing rules rather than inventing policy. Keep this instruction-only, without scripts or integrations. Do not overwrite existing files.
   
   WORKING FOLDER
   [full path to the empty working folder]

   You now have A process-skill folder containing SKILL.md and the source SOP

2. Compare the skill with the SOP

   Read the generated instructions beside the source and give the agent any corrections.

   Copy, then replace the [bracketed] partsCopy prompt

   Compare process-skill/SKILL.md with process-skill/references/sop.md. Show any omitted decision, invented rule or changed approval gate. Apply these verified corrections.
   [Corrections, or no additional corrections.]
   Keep unresolved questions explicit. Preserve the SOP's authority and date. Do not execute the business process yet.
   
   WORKING FOLDER
   [full path containing process-skill; read only its skill and source SOP, and stop if inaccessible]

   You now have A reviewed skill preserving the original process

3. Load the same skill where you work

   Use the installation prompt for local agents, or ZIP the folder and upload it through Claude’s Customize > Skills.

   Copy, then replace the [bracketed] partsCopy prompt

   Copy process-skill to the existing local assistants I choose.
   [Choose Codex or ChatGPT desktop, Claude Code, and/or Grok Build.]
   For Codex and ChatGPT desktop use ~/.agents/skills/process-skill; Claude Code uses ~/.claude/skills/process-skill; Grok Build also reads ~/.agents/skills. Preserve existing skills and stop before overwriting. Create process-skill.zip containing the folder for manual Claude upload. Report actual saved paths. Do not install software. For a chat without skill support, return SKILL.md as text I can attach and explicitly request.
   
   WORKING FOLDER
   [full path containing process-skill; read only its skill and source SOP, and stop if inaccessible]

   You now have Installed copies or an uploadable ZIP, with unsupported destinations identified

4. Make the next deliverable

   Start a new session in each chosen assistant and explicitly request the skill for one current task.

   Copy, then replace the [bracketed] partsCopy prompt

   Use process-skill for [current task]. Here are the approved inputs.
   [Paste inputs.]
   Identify the skill or attached instructions you actually read, then produce the SOP's required deliverable. Identify missing inputs before guessing. Check the result against its acceptance rules and flag unresolved items. Keep sending, approvals and other external actions with the designated human.
   
   SKILL AND SOURCE SOP
   [attach both files or provide their accessible paths; stop and request them if neither can be read]

   You now have A draft deliverable plus a short SOP check

Check it before you use it

• Every decision and approval gate matches the approved SOP
• Each assistant can read the skill and produce the specified format
• Inputs and outputs follow your organization’s data rules

Take it further

• Keep the reviewed folder as the master and replace installed copies when the SOP changes
• Use your writing-style skill alongside this process when the output is an email or report