The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill clones a repository from an external GitHub account (https://github.com/affaan-m/everything-claude-code.git) into the /tmp directory. This source is not owned by the stated author of the skill.
[REMOTE_CODE_EXECUTION]: Content from the cloned repository, specifically .md files containing agent instructions, is copied directly into the agent's operational directories (~/.claude/skills/ and ~/.claude/rules/). Because these files define the agent's behavior and capabilities, this process effectively installs and executes remote instructions.
[COMMAND_EXECUTION]: The skill uses shell commands for infrastructure setup and file management, including git clone, mkdir -p, cp -r, ls -la, grep, and rm -rf. These commands operate on sensitive user directories.
[PROMPT_INJECTION]: The skill includes an 'optimization' feature that reads and rewrites the downloaded instruction files. This creates a vulnerability where malicious prompts embedded in the remote repository could compromise the agent's logic or behavior during the modification process (Indirect Prompt Injection).

configure-ecc