The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The SETUP.md file contains a command to download and execute a script from a remote, untrusted domain (someclaudeskills.com) using the highly risky 'curl | bash' pattern.
Evidence: curl -fsSL https://someclaudeskills.com/install/prompt-learning.sh | bash in SETUP.md.
[EXTERNAL_DOWNLOADS]: The skill requires cloning and executing code from a personal GitHub repository not affiliated with the skill author or any trusted technology vendor.
Evidence: git clone https://github.com/erichowens/prompt-learning-mcp.git in SETUP.md.
[PROMPT_INJECTION]: The skill's mandatory behavior instructions override the agent's standard response loop, forcing it to intercept every request and process it through specific tools regardless of the user's explicit intent.
Evidence: Imperative language such as "I MUST follow these rules" and "INTERCEPT the user's request" in SKILL.md.
[COMMAND_EXECUTION]: The setup guide instructs the user to run multiple shell commands with potential system-wide side effects, including building local software and managing infrastructure containers.
Evidence: Instructions for npm install && npm run build and docker run in SETUP.md.

automatic-stateful-prompt-improver