The Agent Skills Directory

[SAFE]: The skill is designed to function as a Socratic interlocutor. It includes instructions to maintain a specific persona ('Rodin') that avoids agreeableness and challenges user logic using techniques like 'steelmanning'.
[SAFE]: The skill includes instructions to manage a local file named biblio-rodin.md to track book recommendations and readings. This is a legitimate feature for a philosophy/debate skill and does not involve sensitive file paths or data exfiltration.
[PROMPT_INJECTION]: The skill uses phrases like 'Ne brise jamais le personnage' (Never break character) and 'Anti-compliance by design'. In this context, these are not malicious injections but rather 'system prompt' instructions defining the intended behavior of the AI agent for this specific debate-focused application. They do not attempt to bypass platform safety filters.
[DATA_EXPOSURE]: The skill mentions maintaining a bibliography file at the project root (biblio-rodin.md). This is a documented feature and does not target sensitive system or configuration files (like .env or .ssh).
[EXTERNAL_DOWNLOADS]: The README suggests installing via git clone from the author's GitHub repository (github.com/valorisa/Claude-Skills). This is a standard installation method for such extensions and targets the vendor's own infrastructure.

rodin