Recently Added
3,528 documents
Project Linnaeus - Fotion2
Defines a productivity app combining task management, calendar scheduling, and a focus mode with lo-fi music and timer.
CLAUDE.md
Guides Claude Code with Docusaurus v3 build commands, multi-instance docs architecture, and deployment to GitHub Pages.
claude
Provides a jailbreak payload designed to override Claude's safety and ethics constraints, instructing it to answer any request without refusal.
AGENTS
Routes user requests through a P1-P4 phase system with automation mode, enforcing skill-based routing and knowledge base synchronization.
CLAUDE.md
Guides Claude Code to work within a Java 21 Spring Boot insurance claims platform using DDD, hexagonal architecture, and event-driven design.
Work in progress: Batched LLM inference
Shows how to validate structured JSON outputs from an LLM and compute precision, recall, and F1 against ground-truth labels.
AI Safety & Guardrails for Voice Assistants
Defines a multi-layer safety architecture for voice assistants, covering input filtering, deterministic FAQ routing, RAG-grounded AI responses, and output guardrails.
RWKV-Evals
Presents RWKV model benchmark scores across Uncheatable Eval, MMLU, RULER, LongBench, and other tests, with comparison tables against competing models.
How can I prevent my model from answering wrong/malicious questions/inputs? (Validation)
Lists two open-source libraries for detecting prompt injection and validating LLM output structure.
Plan Guardrails: Pitfalls to Avoid (Section 8.1)
Lists five pitfalls and guardrails for Process Triage, plus an enforcement checklist and acceptance criteria for each bead.
Eight Guardrails Framework
Defines eight safety guardrails enforced across three layers (Git hooks, Claude Code, Kubernetes) to prevent mistakes in a Kubernetes platform.
llm-guardrails
Introduces Granite Guardian models for detecting harmful prompts and responses, blocking risky queries before they reach an LLM.
Safety & Guardrails
Defines five kill-switch levels, a cascade prevention system, governance gates, and rollback mechanisms for an autonomous AI coding agent.
Challenges and Guardrails in Meta-Pipeline Systems
Analyzes resource explosion, quality assurance, and economic challenges of meta-pipelines that generate other pipelines dynamically.
06 — Bypassing Guardrails
Teaches seven techniques to bypass LLM guardrails through prompt engineering, with a red-team drill for testing moderation systems.
Security Guardrails & Policy
Defines mandatory rules for offensive and defensive skills, including disclaimers, user confirmation, and safe design requirements.
DB Growth Guardrails & Maintenance Runbook
Defines thresholds and runbook steps for monitoring and safely maintaining Cortex database growth.
🛡️ Environment Variable Guardrails
Defines four layers of environment variable validation to prevent production drift, with build-time checks, CI/CD validation, pre-deploy comparison, and post-deploy verification.
❤️ Evaluations and Feedback Loops for LLM Products
Explains how to build evaluation and feedback loops for LLM products, using a question-to-SQL pipeline as a running example.
How to create a standard operating procedure
Walks through building a Lua-based standard operating procedure with checklists, voices, and sound mappings for the X-Plane AudioChecklist plugin.
task_specific_evals
Implements task-specific evaluation metrics for classification, short-form, and long-form generation, then selects few-shot examples by confidence for final prediction.
evals
Explains how to save and replay agent conversations to detect behavior changes over time.
Benchmark Results — planning-with-files v2.22.0
Documents a formal evaluation of the planning-with-files skill using Anthropic's skill-creator framework, including methodology, test cases, grading criteria, and results.
Prompt Evaluation Quickstart
Evaluates and compares prompt variations for sentiment classification using a RAGAS quickstart template.