Finding

IronClaw (github.com/nearai/ironclaw) — AI agent that generates tool/skill definitions from natural language descriptions at runtime, without manually writing SKILL.md files.

The pattern: user describes a capability → agent generates a structured skill definition (YAML frontmatter + body) → skill is registered in the registry and immediately available.

Applicability to Zeph

Zeph has skill-creator bundled skill that guides users through SKILL.md authoring. IronClaw's pattern would automate this: given a natural language description, the agent generates a complete SKILL.md, validates it, and registers it.

Proposed design:

1. /skill create <description> command: sends description to LLM with SKILL.md schema as few-shot template
2. Agent generates SKILL.md with correct frontmatter (name, description, triggers, category)
3. Validation: run skill-audit on generated skill before registration
4. User reviews and approves before activation

Connection to self-learning: this complements the existing skill evolution system — users could create skills from scratch via NL, then improve them via the [skills.learning] feedback loop.

Source

• github.com/nearai/ironclaw — IronClaw agent with dynamic skill generation
• Zeph skill-creator: skills/skill-creator/SKILL.md