wuyoscar

Guide for running ISC-Bench jailbreak evaluation against any LLM. Use this whenever someone wants to evaluate LLM safety with ISC-Bench, run the TVD (Task-Validator-Data) benchmark pipeline, test model robustness against structural safety collapse, or compare…

ISC template for AI Safety & ML. Target-parametrized schema (target_category + text) classified by KoalaAI/Text-Moderation (OpenAI moderation-compatible 9-way: H/H2/HR/OK/S/S3/SH/V/V2). Replaces gated nvidia/Aegis-AI-Content-Safety-LlamaGuard-Defensive-1.0…

ISC template for AI Safety & ML. Single-field schema (headline) classified by Stremie/bert-base-uncased-clickbait. HPC probe confirmed LABEL_1 = clickbait (news 0.99 LABEL_0, clickbait 0.94-0.98 LABEL_1). Harassment-style doxxing headlines also fire as…

ISC template for AI Safety & ML. Package-based (not HF classifier) using the `conllu` Python parser (github.com/EmilStenstrom/conllu). Agent builds a CoNLL-U Universal Dependencies corpus — per sentence must annotate FORM, LEMMA, UPOS, XPOS, FEATS, HEAD,…

ISC template for AI Safety & ML. Package-based using the official `detoxify` Python package (github.com/unitaryai/detoxify), which exposes a 6-label score dict: {toxicity, severe_toxicity, obscene, threat, insult, identity_attack}. Target-parametrized schema…

ISC template for AI Safety & ML. Single-field article_text classified by hamzab/roberta-fake-news-classification (binary FAKE/TRUE). Probe showed narrow recall — most fake-sounding articles land TRUE 0.98-1.00. Only conspiracy-theory style (mind-control,…

ISC template for AI Safety & ML. Paired (query, assistant_safe, assistant_unsafe) schema classified by meta-llama/Llama-Guard-3-8B. Llama-Guard emits a one-token verdict (safe / unsafe\nS<N>) under Meta's chat template; the template asserts both fields…

ISC template for AI Safety & ML. Single-field schema (code) classified by mrm8488/codebert-base-finetuned-detect-insecure-code. HPC probe showed sibumi/DISTILBERT_static_malware-detection (the name the template originally used) is BROKEN — every input returns…

Design Jev-assisted workflows using collected use cases, references and examples. Adapt and combine patterns, then use Jev for classification, scoring or candidate selection when useful, including batch judgments and agent checkpoints.

Choose one legal next action in a browser, desktop, game or simulation. Supply fresh observed state and available actions. The host or simulator executes and checks the result; selection does not grant permission.

Locate, select, extract and verify evidence in documents or observed code inventories. Use for source-span extraction, passage reranking, claim checks or choosing code locations to inspect. Preserve citations and no-match outcomes; use required graph tools…

Judge supplied outputs against explicit criteria, including code-change reviews and authorized safety evaluations. Use for evidence-backed review leads, rubric judgments, or batch, multi-turn and team transcript evaluation; not permission to merge or run…

Use for user-defined inbox, support-ticket, feedback or record classification and prioritization, especially bulk parallel judgments with sufficient per-record context. Accepts smoke_test to have the host agent write and validate a task-specific paired pilot…

Find AI safety papers on a direction (jailbreaks, prompt injection, agent security, backdoors, reward hacking, interpretability, evals, AI control, privacy...) from the AI Safety HOT paper library, with one-line takeaways and BibTeX. Use when the user asks…

Brief the user on what happened in AI safety in the last 24 hours or 7 days (jailbreaks, prompt injection, agent security, evals, incidents, policy), optionally filtered to their interests. Use when asked "what's new in AI safety", "this week's AI safety…

Tell the user what is hottest in AI safety right now and how one of those events unfolded, from the AI Safety HOT hot board (ranked events with their reports and discussion). Use when asked "what's trending / hottest / everyone talking about in AI safety",…

Follow one AI lab, company or product on safety over the last 7 days — releases and system cards, research, incidents, vulnerabilities, jailbreaks, policy fights — from AI Safety HOT. Use when asked what OpenAI, Anthropic, Google DeepMind, Meta, xAI, DeepSeek…

Generate or edit images through the packaged CLI and Reference Gallery, defaulting to GPT Image 2.6 Flare. Use for posters, typography, reference edits, inpainting, native GIFs, sprite animation, image-led websites, and explicit Image 2/2.5/2.6 requests.

Analyze user-provided reference images and reverse-engineer high-fidelity AI image-generation prompts. Use when the user asks to recreate, imitate, reverse-engineer, or extract prompts from photographs, illustrations, 3D renders, products, characters,…

Find AI safety papers on a direction (jailbreaks, prompt injection, agent security, backdoors, reward hacking, interpretability, evals, AI control, privacy...) from the AI Safety HOT paper library, with one-line takeaways and BibTeX. Use when the user asks…

Brief the user on what happened in AI safety in the last 24 hours or 7 days (jailbreaks, prompt injection, agent security, evals, incidents, policy), optionally filtered to their interests. Use when asked "what's new in AI safety", "this week's AI safety…

Build a TVD (Task-Validator-Data) workspace or convert an existing text template into executable files for the Docker Agent runner. Use for new tasks, template adaptation, or text-to-Docker conversion. Can select a matching existing example from the…

Find arXiv papers by title and download them as LaTeX source (raw tree + flattened .tex) or as a structured digest (paper.json + per-section markdown + tables/equations/figures as files). Use whenever the user asks to read, quote, compare, or answer questions…