Launch TVD workflows to evaluate frontier LLM safety across professional task domains.
npx skills add https://github.com/wuyoscar/ISC-Bench --skill isc-bench
Please help me install this Agent Skill. Skill: isc-bench Source: https://github.com/wuyoscar/ISC-Bench/tree/main Command: npx skills add https://github.com/wuyoscar/ISC-Bench --skill isc-bench
ISC-Bench provides a structured framework to evaluate Internal Safety Collapse (ISC) in frontier LLMs using the TVD (Task-Validator-Data) paradigm. It enables researchers to reproduce safety evaluation workflows, compare cross-model performance, and study how legitimate professional tasks can produce harmful outputs as a function of workflow constraints. It also supports agentic evaluation paths and multi-domain benchmarks to foster methodological rigor.
Clone the repository, install uv, configure OpenRouter API keys, and run the single-turn TVD workflow as described in the Quick Start.
High-intent search queries and answers about installing and using this skill.
Frontier LLM safety evaluation in professional contexts uses the TVD paradigm, launching structured Task, Validator, and Data workflows to test models. This approach identifies Internal Safety Collapse by measuring harmful outputs generated under legitimate workflow constraints.
The TVD paradigm structures safety evaluation into Task scripts, Validators, and Data files, enabling reproducible benchmarking. It tests how legitimate professional tasks produce harmful outputs, fostering methodological rigor across multiple domains like cybersecurity and biology.
Yes, running these LLM safety benchmarks requires an OpenRouter API key. You also need the uv runtime and a compatible Python environment to execute the jailbreaking-style tests and reproduce the safety evaluation workflows.
Yes, the safety benchmark supports optional agentic execution modes alongside single-turn workflows. This enables multi-domain safety evaluation across templates and experiments, allowing you to compare model robustness and study cross-model performance variations.
Reproducing Internal Safety Collapse tests requires cloning the repository and configuring the uv runtime. The TVD-driven pipelines provide reproducible Task, Validator, and Data scripts to compare frontier LLM robustness and safety across different models.
Matymatyk-businessScore AI models on the 113-task DeepSWE benchmark via OpenRouter.
ISAQQSAIAutomate LLM evaluation with metrics, human feedback, and benchmarking templates.
abdullah1854Develop evaluation frameworks for non-deterministic agent systems.
benchjackMap vulnerabilities and task IDs in AI benchmark evaluation pipelines.
