The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill initiates a git clone from a non-trusted GitHub repository (https://github.com/affaan-m/everything-claude-code.git) into the /tmp directory. It uses the downloaded content as the source for all installation operations.
[REMOTE_CODE_EXECUTION]: Content from the unverified repository is copied into active skill and rule directories (~/.claude/skills/ and ~/.claude/rules/). Because AI skills can contain shell commands and system instructions, this mechanism facilitates the persistent execution of remote code within the user's local environment.
[COMMAND_EXECUTION]: The skill uses a wide range of shell commands to perform its operations, including rm -rf, git clone, mkdir -p, cp -r, ls -la, and grep. While these are used for installation and verification, they operate on directories containing agent configurations.
[INDIRECT_PROMPT_INJECTION]: During the 'Optimization' phase in Step 5, the skill reads and programmatically modifies installed SKILL.md files based on user input and project context. This creates a surface where the agent processes and rewrites its own instructions based on potentially malicious data from the external source or project files.

configure-ecc