ToolShield — Development skill for Claude Code
[ICML 2026] Official implementation for paper "Unsafer in Many Turns: Benchmarking and Defending Multi-Turn Safety Risks in Tool-Using Agents".
How to install ToolShield
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open CHATS-lab/ToolShield and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What ToolShield does
[ICML 2026] Official implementation for paper "Unsafer in Many Turns: Benchmarking and Defending Multi-Turn Safety Risks in Tool-Using Agents".
Alternatives in Development
- Understand Anything — Claude Code skills that turn any codebase into an interactive knowledge graph you can explore, search, and ask 2.5k ★
- Vox Collage Broll — Turn one spoken line into a paper-collage explainer video — Claude Code skill + MaxFusion MCP 198 ★
- Contract Review Skill — Review legal contracts for risks, extract key terms, and suggest redlines 192 ★
README
[ICML 2026] ToolShield: A Package to Guard Your Agent
Accepted to the ICML 2026 main conference · 🏆 additionally selected as a Spotlight at the AI4GOOD Workshop @ ICML 2026
[](https://pypi.org/project/toolshield/) [](https://arxiv.org/abs/2602.13379) [](https://unsafer-in-many-turns.github.io) [](LICENSE) [](https://huggingface.co/datasets/CHATS-Lab/MT-AgentRisk) [](https://pepy.tech/projects/toolshield)
Related Skills
Agent Bazaar Implementation
An implementation of the paper: "Agent Bazaar: Enabling Economic Alignment in Multi-Agent Marketplaces" with S
JiegeOSBench
Benchmarking a single‑prompt Rust OS kernel implementation, comparing Claude Code Fable, Opus, and Sonnet.
Clickup Report Skill
Claude Code skill: turn your ClickUp tickets into a meeting-ready sprint report (by sprint → status, with tick
Product Safety Scan Skill
A Claude skill that turns photos of a product label into a sourced verdict: ingredient flags against a list yo
Sleeper Agents Paper
Contains random samples referenced in the paper "Sleeper Agents: Training Robustly Deceptive LLMs that Persist
Phase Breakdown
Execute a development phase using native Claude Code sub-agents for parallel implementation. Supports 33 langu
Related Agents
Forge Researcher
Multi-source research agent that investigates best practices, official documentation, and academic literature
Ticket Planner
Turn a software request and current code into a scoped, executable plan with explicit assumptions, dependencie
Feos Discovery
FEOS S1 Discovery. Use to turn a vision or problem statement into a research pack and risk register before req