Overall risk is driven primarily by a destructive local deletion path in skills.sh. Documentation contains examples of remote-code execution and unpinned installs (supply-chain concern), but there is no evidence of the agent performing those actions or exfiltrating data. Prompt-injection and obfuscation/evasion are not evidenced in the inspectable files.
remove_skill() uses rm -rf "$PROJECT_SKILL_DIR" (derived from "$HOME/.agents/skills/$PROJECT_SKILL_NAME") without additional safety checks.
Includes a remote-code execution example: "curl -fsSL https://cli.inference.sh | sh && infsh login".; Includes @latest-style installation guidance (e.g., "npx skills add marktantongco/aiskills-photog@latest"), increasing version drift risk.
No description provided
npx skillscat add marktantongco/ai-visual-synthesis
Install via the SkillsCat registry.