Agent Skills: skillkit-help

Routing

Detect which path the user needs and jump directly to it.

| User says | Route | |-----------|-------| | "what are skills", "how do skills work", "explain", "understand", "not sure", "should I" | → Path A | | "validate", "check", "review my skill", "is this good" | → Path B | | "ready to build", "let's create", "make a skill" | Tell them: "You're ready — invoke /skillkit directly to start building." | | Ambiguous | Ask: "Do you want to (A) understand how skills work, (B) validate an existing skill, or are you ready to build (invoke /skillkit)?" |

Path A: Understand How Skills Work

Goal: Build a mental model of skills — what they are, when to use them, and whether you actually need one — before starting to build.

Step 1 — Why skills exist

Load and read in full: knowledge/foundation/01-why-skills-exist.md

Summarize for the user: skills are reusable prompt-time instructions that extend your agent's behavior for specific tasks. They live in ~/.claude/skills/ and are invoked via /skill-name.

Step 2 — Skills vs subagents

Load and read: knowledge/foundation/02-skills-vs-subagents-comparison.md

Explain the difference with a concrete example:

• Skill: "When I type /review-pr, load these code review instructions"
• Subagent: "Spin up a separate agent with browser tools to scrape and summarize a URL"

Step 3 — Decision framework

Load and read: knowledge/foundation/03-skills-vs-subagents-decision-tree.md

Walk the user through the decision tree for their specific use case.

Step 4 — Platform constraints

Load and read: knowledge/foundation/06-platform-constraints.md

Cover the key rules: frontmatter requirements, size limits, trigger conditions.

Step 5 — Hand off to the builder

Tell the user: "You now have enough context to start building. Invoke /skillkit — it will guide you through the full creation workflow."

Path B: Validate an Existing Skill

Goal: Check an existing skill for quality issues before sharing or publishing.

Step 1 — Load validation standards

Load and read in full: knowledge/application/12-testing-and-validation.md

Step 2 — Run the checklist

Ask the user to share their SKILL.md content or path. Then check:

• [ ] Frontmatter: name, description, category all present
• [ ] Description has a clear trigger (when to invoke it)
• [ ] At least one concrete usage example in description or body
• [ ] No hardcoded secrets, API keys, or PII
• [ ] SKILL.md is under 500 lines (if over, recommend splitting)
• [ ] Sections are clearly delimited with ## headings
• [ ] Invoke in Claude Code: does it fire correctly?

Report findings: pass/fail per item, specific fix for each failure.