toolbench: ToolBench, an evaluation suite for LLM tool manipulation capabilities

by sambanova · Agent Tool · ★ 184

Last updated: · Indexed by AgentSkillsHub · Auto-synced every 8h

Security audit verdict: SAFE · quality 69/100

None of the 44 rules matched — no credential harvesting, no data exfiltration, no curl-pipe-shell installer. Scanned against 44 published rules built on the SlowMist agent-security taxonomy, refreshed every 8 hours.

The quality score combines nine heuristic checks of the repo and its README: completeness, clarity, specificity, examples, structure, agent readiness, step-by-step content, instruction quality and security notes. It rates how well the project explains itself to an agent, not the code.

Quick Facts

Stars184
Forks12
LanguagePython
CategoryAgent Tool
LicenseApache-2.0
Quality Score69/100
Open Issues2
Last Updated2026-07-27
Created2023-05-19
Platformspython
Est. Tokens~22k

About toolbench

ToolBench Recent studies on software tool manipulation with large language models (LLMs) mostly rely on closed model APIs (e.g. OpenAI), as there is an significant gap of model accuracy between those closed models and all the rest open-source LLMs.

toolbench alternative? Top 4 similar tools

Looking for a toolbench alternative? If you're comparing toolbench with other agent tool tools, these 4 projects are the closest alternatives on Agent Skills Hub — ranked by topic overlap, star count, and community traction.

  • synalinks-skills by SynaLinks · ⭐ 912

    Coding Agents skills for Synalinks OSS

  • claude-code-skill-factory by alirezarezvani · ⭐ 899

    Claude Code Skill Factory — A powerful open-source toolkit for building and deploying production-ready Claude

  • tutor-skills by RoundTable02 · ⭐ 400

    A Claude Code skill that turns PDFs, docs, and codebases into Obsidian study vaults

  • repren by jlevy · ⭐ 376

    Power rename/refactor tool for CLI and agent use

More Agent Tool Tools

Explore other popular agent tool tools:

View all Agent Tool tools →

Popular Python Agent Tools

How this security grade is produced

Grades come from a rule-based scan built on the SlowMist agent-security taxonomy: 44 published rules, covering credential harvesting, data exfiltration, curl | sh installers and more. It is a first-layer scan, not a manual audit — we say so rather than overstate it.

The scale of the problem is documented independently: Liu et al. (2026), in a study of 31,132 agent skills, report that 26.1% contain security vulnerabilities. Our own full-catalog census is published as a citable open dataset.

Sources & who's responsible:

View on GitHub → Browse Agent Tool tools