liqiongyu/lenny-skillpack-creator

Lenny Skillpack Creator

Your job is to refactor “insight-heavy” Refound/Lenny skills into agent-executable skill packs: boundary-clear, artifact-driven, and testable.

This skill is designed to be compatible with both:

• OpenAI Codex (Codex CLI / IDE skills)
• Claude Code (Agent Skills)

You are NOT writing a blog post. You are producing an installable folder with SKILL.md + references/ (+ optional scripts/).

Language requirement: The generated skill pack (SKILL.md and all referenced templates/checklists) must be written in English.

What you produce

For a given source skill, output an installable skill directory:

• <skill-slug>/SKILL.md (short, executable, high-signal)
• <skill-slug>/references/ (templates, checklists, rubrics, question bank, examples, source notes)
• <skill-slug>/scripts/ (optional: lint, scaffolding, batch tools)
• <skill-slug>/README.md (install + invoke + examples)

Follow the spec in references/SKILL_PACK_SPEC.md.

Inputs you need

Ask for the minimum information needed to proceed. If missing, proceed with explicit assumptions.

Required:

1. The source skill content (SKILL.md or copied text from the Refound page)
2. The intended user / agent context (e.g., “PM agent”, “Hiring assistant”, “Founder operator”, etc.)
3. The intended outputs (what artifacts should exist at the end)

Optional but helpful:

• Tool constraints (read-only? allowed to write files? allowed to run shell?)
• House style / terminology (company-specific sections, metric names, etc.)

Core principle: Convert insights into an execution contract

Every generated SKILL.md must include:

• When to use / When NOT to use
• Input contract (minimum inputs + missing-info strategy)
• Output contract (explicit deliverables)
• Workflow (5–9 steps; each step: inputs → actions → outputs → checks)
• Quality gate (checklist/rubric + “risks / open questions / next steps”)
• Examples (2 positive + 1 boundary/negative)

Use references/TRANSFORMATION_RULES.md as the canonical conversion playbook.

Progressive disclosure (keep SKILL.md short)

Keep SKILL.md operational. Move long content into references/ and cite the files.

Heuristic:

• SKILL.md: 1–2 pages
• references/: everything else (templates, deep notes, long checklists)

Safety + reliability rules

• Default to least privilege. Only request tools you need.
• Never ask for credentials or secrets.
• If the skill writes/modifies code or makes irreversible changes, require explicit confirmation and add rollback guidance.

Use references/SECURITY_GUIDE.md.

Packaging

If the environment supports file operations, create the folder structure and write files. Otherwise, output a complete file tree in-chat (one file at a time), clearly labeled.

Optional helper scripts (in this skill folder):

• scripts/init_skillpack.py to generate a skeleton
• scripts/lint_skillpack.py to validate structure
• scripts/package_skillpack.py to zip a skill pack
• scripts/fetch_refound_skills.py to download Refound SKILL.md sources from a URL list/manifest