skill-creator-pro

name	skill-creator-pro
description	Design, create, review, and iteratively improve high-quality AI skills with strong trigger definitions, progressive disclosure, reusable scripts/references/assets planning, validation rules, and anti-pattern avoidance. Use when asked to create a new skill, upgrade an existing skill, turn a repeated workflow into a reusable skill, review skill quality, or define skill design best practices.

Skill Creator Pro

Goal

Design and refine production-grade skills that trigger correctly, stay lean in context, execute predictably, and improve after real usage. Automatically recommend the best design paradigm for each new skill.

Operating Modes

• Create: recommend paradigm → define boundary → plan resources → generate structure → validate.
• Review: inspect an existing skill for structural defects, content-quality issues, routing weaknesses, and missing validation.
• Upgrade: review first, then apply the smallest changes that materially improve trigger quality, clarity, reuse, and verification.

Create Workflow (Five Phases)

Phase 0 — Paradigm Selection Run scripts/paradigm_recommender.py --description "user's description" to get a paradigm recommendation. If confidence ≥ 0.7, present the recommendation and ask user to confirm. If confidence < 0.7, ask clarifying questions before recommending. Read references/paradigms/{paradigm}.md for the confirmed paradigm's best practices.

Phase 1 — Boundary Definition Answer four questions before writing anything:

• What repeated problem does this skill solve?
• What user requests should trigger it?
• What nearby requests should not trigger it?
• What counts as success?

Phase 2 — Resource Planning Based on paradigm, decide what goes into scripts/, references/, and assets/. Use the paradigm-specific resource heuristics from references/paradigms/{paradigm}.md.

Phase 3 — Generate Skill Scaffold with scripts/init_skill_pro.py <name> --path <dir> --paradigm <paradigm>. Then replace all template wording with real content from Phase 1 and 2.

Phase 4 — Validate Run scripts/review_skill.py <path>. Fix any high-severity findings. Test at least one realistic usage path end to end.

Review and Upgrade Workflow

1. Run scripts/review_skill.py <path/to/skill>.
2. Separate findings: hard structural failures → routing weaknesses → content issues → polish.
3. Run scripts/paradigm_recommender.py --skill-path <path> to check paradigm alignment.
4. Fix trigger and routing before expanding scope.
5. Hand off to skill-benchmark when user wants effectiveness evidence.

Paradigm Quick Reference

Paradigm	Core Job	Main Risk	Key Directory	Share
Operator	Execute toolchain reliably	Execution failure	scripts/	28%
Navigator	Route to right information	Wrong guidance	references/	19%
Architect	Produce reusable systems	Non-reusable output	assets/	17%
Partner	Structure collaboration	Misunderstood intent	references/	14%
Orchestrator	Coordinate multi-tool/agent	Coordination chaos	references/	13%
Scout	Inspect before acting	Wrong assumptions	references/	6%
Philosopher	Establish governing principles	Inconsistent behavior	references/	3%

Paradigm Decision Tree

Identify the main risk, then match:

• Execution fails or results inconsistent → Operator
• User gets wrong information → Navigator
• Output is not reusable or systematic → Architect
• User intent misunderstood → Partner
• Multi-tool coordination breaks → Orchestrator
• Acting on wrong assumptions → Scout
• Behavior lacks principled consistency → Philosopher

When unclear, ask: Does it need scripts for determinism? (Operator) Large knowledge base? (Navigator) User confirmations? (Partner) Environment recon? (Scout) Reusable templates? (Architect) Multi-agent handoff? (Orchestrator) Constitutional rules? (Philosopher)

Boundary First

Before creating or revising a skill, answer:

• What repeated problem does this skill solve?
• What user requests should trigger it?
• What nearby requests should not trigger it?
• What counts as success for the skill user? If answers are fuzzy, use references/templates/request-templates.md to collect missing info.

Design Rules

• Prefer narrow and strong over broad and vague.
• Choose the paradigm before choosing the structure.
• Frontmatter = Routing Layer. SKILL.md = Control Layer. scripts/ references/ assets/ = Execution Support.
• Keep SKILL.md focused on workflow, decision points, constraints, validation, and resource routing.
• Do not duplicate information across SKILL.md and references/.
• Do not explain basics the model already knows.
• Do not create empty directories.

Trigger Quality

• name: lowercase hyphen-case.
• description: capability + invocation context + representative tasks/objects.
• Include "Use when..." phrasing.
• Keep trigger logic in frontmatter, not body.

Recommended SKILL.md Shape

---
name: my-skill
description: [capability + "Use when..." + trigger contexts]
---
# My Skill
## Goal
## Workflow
## Decision Tree
## Constraints
## Validation
## Resources

Validation Standard

• frontmatter exists and is valid YAML
• name and description are present and correct
• directory structure matches paradigm needs
• agents/openai.yaml reflects the skill
• at least one realistic usage path works end to end

Anti-Patterns

• Writing a tutorial instead of an execution guide
• Making the skill broad before making it strong
• Mixing trigger rules into body instead of frontmatter
• Copying large reference content into SKILL.md
• Creating empty or decorative directories
• Declaring success without a validation path
• Choosing wrong paradigm (e.g., treating Operator as Navigator)

Resources

• scripts/paradigm_recommender.py: recommend paradigm from description or existing skill
• scripts/init_skill_pro.py: scaffold skill structure with paradigm-specific templates
• scripts/review_skill.py: review an existing skill for structural and content issues
• references/paradigms/: seven paradigm best-practice files (operator, navigator, architect, partner, orchestrator, scout, philosopher)
• references/paradigms/hybrid-patterns.md: when and how to combine paradigms
• references/core/design-playbook.md: boundary, scope, and build order
• references/core/not-to-do-red-lines.md: anti-patterns and failure classes
• references/core/examples.md: strong and weak skill patterns
• references/advanced/constitution.md: 10 constitutional rules
• references/advanced/content-review.md: content quality and routing strength rubric
• references/advanced/remediation-playbook.md: findings to fixes mapping
• references/advanced/evaluation-handoff.md: when to hand off to skill-benchmark
• references/templates/request-templates.md: collect missing info for new skills
• references/validation/checklists.md: pre-release checklists
• references/validation/skill-review-scorecard.md: five-dimension scoring
• assets/skill-templates/: paradigm-specific SKILL.md starter templates