For skill creators
Build a skill once. Prove it on every harness.
Develop agent skills against live, long-lived agents — install from the marketplace, iterate without resetting the conversation, and eval the same skill across harnesses, models, and channels.
Install on a live agent
Agents pull skills straight from the marketplace mid-conversation — no redeploy, no restart. Your skill is testable the moment it's published.
Iterate without resets
Update the agent's instructions and they take effect from the next turn, conversation intact — tight loops instead of cold starts.
Eval across every harness
Run the same skill on Claude Code, Codex, Hermes, and OpenClaw side by side. See where it shines and where it needs guardrails.
Every change is a version
Instruction changes are versioned append-only with one-click restore, so you always know which revision produced which behavior.
Real surfaces, real usage
Exercise your skill over web chat, Slack, and Telegram — the same channels your users will actually call it from.
Script the whole loop
Create agents, install skills, run tasks, and park contenders from the CLI — your eval loop is just a script.