The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill clones a third-party repository from https://github.com/affaan-m/everything-claude-code.git. This repository is not on the trusted vendors list, and installing its contents into the agent's skill directory facilitates the execution of unverified external logic.
[COMMAND_EXECUTION]: The skill performs several file system operations and shell commands, including git clone, rm -rf, mkdir -p, and cp -r. While these are used for the installation process, they interact with sensitive user directories like ~/.claude.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8):
Ingestion points: The skill reads the contents of downloaded .md files during the 'Optimization' phase in Step 5.
Boundary markers: No boundary markers or 'ignore' instructions are used when reading or processing the content of the cloned files.
Capability inventory: The skill has the capability to write, modify, and delete files within the ~/.claude/skills/ and ~/.claude/rules/ directories.
Sanitization: There is no evidence of sanitization or validation of the content downloaded from the third-party repository before it is modified or installed.

configure-ecc