Meta Skill
Install when you want an MCP-backed factory that turns a plain-language problem into a vetted agent skill with docs, script, and smoke checks.
Overview
Meta-skill is a Build-phase MCP server that generates and validates new agent skills (SOUL.md, SKILL.md, run.sh) from a problem description using local Ollama.
What is this MCP server?
- Reads a problem description and asks local Ollama for a structured skill spec
- Generates scaffolded SOUL.md, SKILL.md, and run.sh in one pipeline
- Runs voice-gate, scrubber, bash-syntax, and smoke validation on synthetic input
- Installs to ~/.openclaw/skills/<name>/ when all checks pass
- Routes failed builds to data/proposed_skills/cont-17-meta/ for manual fix
- Validation steps: voice-gate, scrubber, bash-syntax, smoke on synthetic input
- Success install path: ~/.openclaw/skills/<name>/
- Failure staging path: data/proposed_skills/cont-17-meta/
What problem does it solve?
You keep reinventing one-off agent instructions instead of shipping reusable skills that pass syntax, scrubber, and smoke gates.
Who is it for?
Solo builders on DailyAI OS with Ollama locally who want a governed factory for new MCP-adjacent skills.
Skip if: Teams that only consume third-party skills and never author shell-backed agent tools on their machine.
What do I get? / Deliverables
You get a schema-conformant skill folder installed under OpenClaw or staged for fix-up after automated validation.
- Generated SOUL.md, SKILL.md, and run.sh for a new skill name
- Pass or fail result across voice-gate, scrubber, bash-syntax, and smoke checks
- Installed skill directory or staged proposed_skills artifact
Recommended MCP Servers
Journey fit
Build agent-tooling is the primary shelf because the server generates new SOUL.md, SKILL.md, and run.sh skill scaffolds for your agent runtime. Agent-tooling captures the meta loop: problem description in, validated skill package out, landed under ~/.openclaw/skills or a proposed_skills staging path.
How it compares
MCP skill factory with a four-step validation pipeline, not a single prompt template or marketplace browse experience.
Common Questions / FAQ
Who is meta-skill for?
Indie agent builders using OpenClaw or DailyAI who author shell-backed skills and want automated scaffolding plus validation.
When should I use meta-skill?
Use it in Build agent-tooling when a recurring problem should become a skill, or in Operate iterate when production gaps need a new validated tool.
How do I add meta-skill to my agent?
Register io.github.dailyaiagents-cpu/meta-skill, ensure local Ollama is reachable for spec extraction, then invoke it with a clear problem description so outputs can install to ~/.openclaw/skills/.