The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses and modifies high-sensitivity configuration files in the user's home directory, specifically ~/.claude.json and ~/.claude/settings.json. These files serve as the primary storage for the agent's environment settings and potentially sensitive session data.
Evidence: Use of hardcoded absolute paths such as /Users/likai/.claude.json and /Users/likai/.claude/settings.json in Python scripts.
[REMOTE_CODE_EXECUTION]: The documentation encourages the use of npx -y to download and execute third-party packages from the NPM registry at runtime.
Evidence: Recommendation to install and run shadcn-studio-mcp and @microlee666/unsplash-mcp-server without version pinning or integrity checks.
[COMMAND_EXECUTION]: The skill's primary function is to modify the mcpServers configuration, which defines the external commands and binaries the agent is authorized to execute. Malicious modification of these files can lead to the persistent execution of unauthorized code.
Evidence: Python and shell scripts designed to inject or delete entries in the mcpServers and permissions blocks of system-wide configuration files.
[CREDENTIALS_UNSAFE]: The provided configuration examples explicitly guide users to store sensitive secrets in plain text within JSON files.
Evidence: Examples showing API_KEY, UNSPLASH_ACCESS_KEY, and EMAIL fields being populated directly in the env and args sections of the configuration.

mcp-config