Skill Architect

Skill Architect

Create, audit, improve, and package agent skills following the Agent Skills open standard. Every skill produced must include a deployment-ready .zip file.

Read references/detail.md before executing. It contains the full platform compatibility matrix, step-by-step creation procedures, frontmatter field reference, quality checklist, packaging commands, and troubleshooting guide.

Triggers

Activate when the user asks to: create a skill, build a skill, package a skill, improve or audit an existing skill, write a SKILL.md, configure frontmatter, structure bundled resources, or distribute a skill.

Mandatory Output Structure

Every skill produced must include:

skill-name/
├── SKILL.md              (required)
├── README.md             (required)
├── references/           (required, even if empty)
├── scripts/              (required, even if empty)
├── assets/               (required, even if empty)
└── skill-name.zip        (required — always, no exceptions)

.zip Packaging Rule

skill-name.zip
└── skill-name/
    ├── SKILL.md
    ├── README.md
    ├── references/
    ├── scripts/
    └── assets/

No .DS_Store, __pycache__, .git, or the .zip itself inside the archive. Always nest inside the skill-name folder — never loose files at root.

The .zip must be produced on every skill creation or update. No exceptions.

Input

• User description of the skill's purpose, or an existing skill folder to audit/improve
• Optional: example invocations, desired trigger phrases, target platform(s)

Output

• skill-name/SKILL.md — frontmatter + instructions
• skill-name/README.md — installation and usage
• skill-name/references/ — supporting documentation
• skill-name/scripts/ — executable code (if needed)
• skill-name/assets/ — templates and output files (if needed)
• skill-name/skill-name.zip — deployment-ready archive

Creation Process (Summary)

1. Capture intent — gather concrete examples of how the skill will be used
2. Plan resources — identify what belongs in scripts/, references/, assets/
3. Initialize directory — mkdir -p skill-name/{references,scripts,assets}
4. Write description — most critical field; must include MANDATORY TRIGGERS: list; max 1024 chars; third person; slightly "pushy"
5. Write SKILL.md body — imperative form; point to reference files, don't duplicate them; under 500 lines
6. Implement bundled resources — scripts with error handling and execute permissions, references under 10K tokens each
7. Write README.md — installation instructions + file listing
8. Validate and package — run scripts/package_skill.py or manual checklist, then zip
9. Iterate — fix skill when Claude struggles; do not fix the prompt

Full procedures for each step: references/detail.md.

Progressive Disclosure Architecture

Scripts execute without loading into context (zero token cost). Only script output consumes tokens.

Bundled Resources Quick Reference

Do not include: standard library docs, credentials, generated/cached files, large datasets, hidden OS files.

Reference Files

Skill Architect v3.0 — April 2026 | agentskills.io open standard