Interactive skill for scaffolding new agent configurations

Summary

Create a DocsClaw skill (or CLI wizard) that interactively scaffolds a complete agent config directory by asking questions about the agent's purpose, suggesting existing skills, and generating all necessary files.

Motivation

Setting up a new agent requires creating several files (system-prompt.txt, agent-card.json, agent-config.yaml, skills/) with the right structure and content. An interactive wizard would lower the barrier to creating new agents and ensure consistency.

Proposed workflow

1. Ask about the agent's purpose and domain (e.g., "document reviewer", "data analyst")
2. Ask about the target LLM provider and model
3. Suggest relevant existing skills from a catalog
4. Ask if additional custom skills are needed
5. Ask about tool requirements (exec, web_fetch, read_file, etc.)
6. Generate the complete config directory:
   • system-prompt.txt — tailored to the stated purpose
   • agent-card.json — with name, description, skills metadata
   • agent-config.yaml — with appropriate tools and loop settings
   • skills/ — copies of selected existing skills
7. Optionally generate ConfigMap YAML via the tooling from #6

Implementation options

• Claude Code skill — a SKILL.md that guides Claude Code through the scaffolding process
• Cobra subcommand — docsclaw init --config-dir ./my-agent with interactive prompts
• Both — the skill for Claude Code users, the subcommand for standalone use

Related

• #6 — Skill-to-ConfigMap conversion tooling (generated config can be immediately converted)
• #7 — Specialized Dockerfiles (wizard could suggest appropriate image flavor)

Start by reviewing the existing configuration files—system-prompt.txt, agent-card.json, agent-config.yaml, and skills/—and the tooling from #6. Compare the Claude Code skill and Cobra subcommand options, then define the interaction and generated outputs before implementing. Done means a new agent directory can be scaffolded consistently, with optional ConfigMap YAML generation.