Creating Skills From Knowledge Folder

Creating Skills from Knowledge Folder

When to use this skill

• When the user provides a folder containing domain knowledge and asks to build new skills from it.
• When the user asks to batch-create or update skills from curated references.
• When skill drafting must follow creating-agent-skills and be audited by reviewing-agent-skills.

Workflow

1. Plan

• Confirm the knowledge folder path and target output path.
• Confirm scope: create new skills, update existing skills, or both.
• Confirm target domains (for example: finance, engineering, legal).

2. Validate

• Build a file inventory (type, size, language, recency).
• Mark unsupported or low-quality inputs.
• Group files by domain intent and likely skill boundaries.

3. Execute

• For each domain group:
  • Propose skill name candidates (gerund, lowercase-hyphen).
  • Draft SKILL.md using creating-agent-skills conventions.
  • Run reviewer checklist from reviewing-agent-skills.
  • Apply fixes and output final skill package.

Instructions

• Treat the folder as source-of-truth evidence, not a prompt dump.
• Do not merge unrelated domains into one oversized skill.
• Prefer multiple small skills over one broad skill.
• Keep each generated SKILL.md focused on:
  • Trigger conditions
  • Deterministic workflow
  • Validation/error handling
• If source material conflicts, preserve both viewpoints and ask user to resolve policy conflicts.

Intake checklist

• Folder path confirmed
• Read permissions confirmed
• Inventory generated
• Domain clusters identified
• Candidate skill map approved

Output contract

For each generated skill, return:

## Generated Skill: <skill-id>
- Source files used: <list>
- Scope boundaries: <inclusions/exclusions>
- Risk notes: <hallucination risk, stale docs, conflicts>
- Audit verdict: <PASS/WARN/FAIL>

Error handling

• If folder is missing or unreadable, stop and request a valid path.
• If files are too large or noisy, propose staged ingestion by subfolder.
• If confidence is low due to sparse sources, produce draft with explicit WARN status.

Resources

• creating-agent-skills
• reviewing-agent-skills