Users find skills online: Claude Skills (SKILL.md packages), prompt recipes, agent toolkits. None of them enter this scaffold verbatim. Larry routes the request; Nolan rules the conversion. A skill is DECOMPOSED into the scaffold's native shapes: | Skill part | Becomes | Where | | --- | --- | --- | | Procedure ("do X then Y") | an SOP, steps marked [JUDGEMENT]/[SCRIPT] | `SOPs/` | | Bundled scripts / code | scripts, red-tested before any SOP points at them | `Scripts/` | | Reference tables, r...
---
type: sop
id: SOP-1012
title: Convert an external skill into scaffold shape
created: 2026-08-28
owner: nolan
skill_name: import-skill
skill_summary: 'Converts an external skill (a SKILL.md package, a prompt recipe, an agent toolkit found online or on disk) into the scaffold''s own shapes: procedures become SOPs, code becomes red-tested Scripts, reference becomes Guidelines, a role gets an agent ruling, and a trigger becomes a generated pointer skill; the user approves the decomposition before any write.'
skill_triggers:
- 'import this skill'
- 'add this skill'
- 'convert this skill'
- 'here is a skill folder'
- 'turn this SKILL.md into scaffold shape'
uses: ["[[SOP-1007-hire-a-new-agent]]", "[[SOP-1011-import-or-align-an-external-agent]]", "[[GL-1004-naming-rules]]", "[[GL-1005-code-vs-instructions]]"]
---
# SOP-1012 Convert an external skill into scaffold shape
Users find skills online: Claude Skills (SKILL.md packages), prompt
recipes, agent toolkits. None of them enter this scaffold verbatim.
Larry routes the request; Nolan rules the conversion. A skill is
DECOMPOSED into the scaffold's native shapes:
| Skill part | Becomes | Where |
| --- | --- | --- |
| Procedure ("do X then Y") | an SOP, steps marked [JUDGEMENT]/[SCRIPT] | `SOPs/` |
| Bundled scripts / code | scripts, red-tested before any SOP points at them | `Scripts/` |
| Reference tables, rules, style guides | a Guideline | `Guidelines/` |
| A role in disguise ("you are a...") | agent ruling via [[SOP-1011-import-or-align-an-external-agent|SOP-1011]] (hire / merge / extract) | `Agents/` |
| Invocation trigger (slash command) | a THIN runtime shim pointing at the SOP | `.claude/skills/<name>/SKILL.md` |
Steps:
1. [SCRIPT] `Scripts/import-inventory.py <skill-folder>` produces the
inventory (shape `skill-package` when a SKILL.md is present, and
every SKILL.md and script listed).
2. [JUDGEMENT] Nolan drafts the decomposition per the table above,
with one line per target file. **The user approves before any
write.**
3. Write the pieces:
- SOPs and Guidelines get the next free number ([[GL-1004-naming-rules|GL-1004]] naming) and
frontmatter per [[GL-1002-frontmatter-conventions|GL-1002]]; foreign prose is rewritten to this
scaffold's shape, never pasted wholesale.
- Scripts land in `Scripts/`, are made ROOT-relative (no absolute
paths from the source machine), and each new guard gets a case in
`run-red-tests.py` and is watched go red ([[GL-1005-code-vs-instructions|GL-1005]] rule 4).
- If the skill must stay invocable by name, add `skill_summary` and
`skill_triggers` (and `skill_prerun` when step 1 is an argument-free
script) to the owning SOP's frontmatter; `Scripts/scaffold-init.py
plan`, then `apply`, writes the `SKILL.md` pointer from them into
`06 AI Team/AI Team Knowledge/Skills/<name>/` and links it into
`.claude/skills/` ([[SOP-1007-hire-a-new-agent|SOP-1007]] step 6b).
The SOP is canonical; the generated file is a pointer. Never write
or edit a `SKILL.md` by hand.
4. Conflicts: a skill instruction that collides with AGENTS.md hard
rules or an existing SOP loses, and the conflict is reported to
the user, never silently resolved.
5. The manifest ([[WS-1004-import-and-convert-external-knowledge|WS-1004]] shape) records source, license note if the
skill carries one, and every file created. Update the touched
agents' bios if their works-with lists grew.
