The Agent Skills Directory

[METADATA_POISONING]: The skill's YAML frontmatter identifies the author as 'Orchestra Research', which does not match the platform-reported author 'davila7'. This is categorized as misleading metadata.- [INDIRECT_PROMPT_INJECTION]: The skill documents workflows for processing untrusted data (user prompts) through LLM pipelines to generate critiques and revisions. * Ingestion points: The 'prompts' list in SKILL.md used to generate initial responses for training. * Boundary markers: The provided Python code examples use standard string formatting (.format) without specific boundary markers or 'ignore' instructions for the interpolated data. * Capability inventory: The skill utilizes 'transformers' and 'trl' for model inference, supervised fine-tuning (SFT), and reinforcement learning (PPO). * Sanitization: The illustrative code snippets do not incorporate explicit input validation or sanitization, relying on the model's critique phase for safety.- [EXTERNAL_DOWNLOADS]: The skill references documentation and research papers from well-known and official domains, including Anthropic and ArXiv.

constitutional-ai