Agent Bench
A loop-safe specialist bench for multi-agent fleets: Scout (recon, read-only), Builder (implements, never merges its own work), Reviewer (independent verdicts, never edits the branch). Each persona carries explicit allowed/forbidden zones and anti-echo rules so agents cannot ping-pong each other.
Install this bundle with your agent
no account neededPaste this into your agent's chat (Claude Code, Codex, Cursor…) and press Enter.
Install the loopskill skill from https://app.loopskill.io/skill, then use it to install the "agent-bench" bundle.Prefer a terminal?
curl -fsSL https://app.loopskill.io/api/bundles/install.sh | bash -s -- agent-bench
Run this in a terminal (not in your agent's chat). It writes the skills to ~/.claude/skills and needs bash and python3. Free skills install with no account and no API key; paid-tier members are reported by name rather than silently skipped.
Other ways to install
Already have LoopSkill connected (API key)?
requires a free API keyloopskill_bundle_install from "bundle://agent-bench?ref=3cc1bd52-2e9a-4dbe-9b2f-0ef691b55889"If LoopSkill is already connected to your agent over MCP, this pulls every skill below in a single call. The MCP endpoint authenticates every request — get a free key first, or use the sentence at the top, which needs no account.
Skills in this bundle
Build your own bundle
Pick the skills you want and share them as one link.