Scaffold

Scaffold — AI Toolkit Generator

Internal utility for the AI Toolkit repository. Creates new agents, new skills, and validates existing plugins against marketplace conventions.

Invocation

This is NOT a plugin skill. It is invoked as a repo-level command:

/scaffold agent <description>
/scaffold skill <description>
/scaffold validate <path>

The input after the action keyword is free-form natural language. The skill parses the intent (agent / skill / validate) from the first word after /scaffold, then treats everything else as the description of what to create or validate.

Workflow Routing

Routing Logic

1. Read the user's input after /scaffold
2. Extract the first significant word to determine intent
3. If ambiguous, ask the user using ask_user:
   • "Do you want to create an agent, a skill, or validate a plugin?"
   • Choices: ["Create an agent", "Create a skill", "Validate a plugin"]
4. Pass the full description text to the selected workflow

How It Works

Agent Creation (/scaffold agent ...)

1. Parses the description to understand the desired agent
2. Asks clarifying questions: target plugin, domain, role, specialization
3. Reads existing agents as reference templates
4. Generates the agent file following established conventions
5. Places it in the correct plugins/<plugin>/agents/ directory
6. Updates affected catalogs (orchestrate.md, AGENTS.md)

Skill Creation (/scaffold skill ...)

1. Parses the description to understand the desired skill
2. Asks clarifying questions: target plugin, purpose, workflows needed
3. Reads existing skills as reference templates
4. Generates SKILL.md + workflow files
5. Places them in plugins/<plugin>/skills/<name>/

Plugin Validation (/scaffold validate ...)

1. Reads the target plugin directory
2. Checks against conventions from CONTRIBUTING.md
3. Validates: required files, naming, size limits, guardrails, consistency
4. Reports issues with severity levels

Principles

Always Confirm Understanding

Before generating anything:

1. Restate what you understood from the user's input
2. Present the plan (what files will be created, where)
3. Ask for confirmation with ask_user
4. Only proceed after explicit approval

Never Overwrite

• If a file already exists at the target path, STOP and ask the user
• Offer: rename, skip, or replace (with explicit confirmation)

Template-Driven

• Always read at least one existing agent/skill as a reference template
• Match the established patterns: frontmatter, XML blocks, section order
• Adapt the template to the new agent/skill's specific role

Incremental Output

• Create files one at a time, confirming each before moving to the next
• After all files are created, offer to update catalogs

Context Files

• See workflows/create-agent.md for agent generation
• See workflows/create-skill.md for skill generation
• See workflows/validate.md for plugin validation
• See ../../CONTRIBUTING.md for marketplace conventions

Guardrails

This skill creates files in the repository structure. Safety rules:

1. Scope: Only writes to plugins/ directories and .github/ — never to system paths
2. No overwrites: Never replaces existing files without explicit user confirmation
3. No execution: Generated agents and skills are documentation — this skill never runs them
4. No system modification: No sudo, no package installs, no network requests
5. Size limits: All generated files must be under 30,000 characters
6. Validation first: Before writing, validate that the target plugin directory exists and follows conventions
7. Dry run option: User can request a preview of what will be generated before writing