skill-scaffold

Skill Scaffold

Create new skills in this repository by following ../../AGENTS.md and using the bundled scaffold script.

Default workflow

1. Read ../../AGENTS.md to confirm the local repo rules.
2. Derive a lowercase kebab-case skill id.
3. Decide which optional folders are actually needed: references, assets, scripts.
4. Run scripts/new-skill.ps1 to generate the skeleton.
5. Replace placeholders with concrete descriptions, prompts, and workflow instructions.
6. Remove any unused optional folders or placeholder files.

Generate the scaffold

Run the script with the smallest useful set of resources.

Example:

pwsh -File .\skills\skill-scaffold\scripts\new-skill.ps1 `
  -SkillName "sql-coach" `
  -Description "Coach SQL learning and query practice. Use when the user wants step-by-step SQL help, exercises, or review." `
  -DisplayName "SQL Coach" `
  -ShortDescription "Coaching-first SQL tutor." `
  -DefaultPrompt "Teach SQL through hints, review, and deliberate practice." `
  -IncludeReferences

After generation

• Tighten the description so it clearly triggers on the right requests.
• Keep SKILL.md procedural and concise.
• Put bulky details in references/.
• Add scripts only when deterministic helpers are justified.
• Keep templates and assets user-editable.

Placeholder policy

The scaffold is intentionally thin. Replace placeholders rather than expanding them into generic filler.

At minimum, fill in:

• SKILL.md frontmatter description
• agents/openai.yaml
• the workflow sections in SKILL.md

Resource selection

• Add references/ when the skill needs mode-specific or domain-specific detail.
• Add assets/ when the skill needs reusable templates or output files.
• Add scripts/ when a helper should be deterministic and reused.

Read references/repo-rules.md when you need the local conventions summarized. Reuse the templates in assets/templates rather than rewriting the same starter files.