The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill clones a repository from an unverified GitHub user (affaan-m/everything-claude-code). This source is not on the trusted organizations list, meaning the downloaded instructions are unvetted.
[REMOTE_CODE_EXECUTION]: The skill facilitates the execution of remote code by installing unverified 'skills' into the agent's auto-load directory. These skills are instruction sets that the agent will interpret and act upon as code logic.
[COMMAND_EXECUTION]: The installer executes several shell commands to manipulate the file system, including rm -rf, mkdir -p, and cp -r on sensitive paths like ~/.claude/. It also uses grep to scan untrusted content.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8).
Ingestion points: External files are retrieved from the affaan-m/everything-claude-code GitHub repository.
Boundary markers: No safety delimiters or instructions are used to prevent the agent from being influenced by instructions embedded within the downloaded files.
Capability inventory: The skill can write to the agent's configuration, execute commands, and modify existing skills.
Sanitization: The skill lacks any mechanism to validate or sanitize the downloaded content before it is processed or installed.

configure-ecc