The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements a caching workflow that uses eval "$formatter_cmd" to execute command strings retrieved from a local JSON file (.claude/cache/format/config.json). This presents a security risk as any modification to the local cache file could lead to arbitrary command execution when the skill is invoked.
[COMMAND_EXECUTION]: The skill uses shell pipes and xargs to process file lists from the filesystem (e.g., git diff --name-only | xargs npx prettier --write). Without the use of null-termination (e.g., git diff -z | xargs -0), this pattern is vulnerable to command injection or unexpected behavior if a repository contains files with maliciously crafted names including shell metacharacters like semicolons or backticks.
[EXTERNAL_DOWNLOADS]: The skill frequently uses npx to execute tools like Prettier, ESLint, and Stylelint. This can result in the automatic download and execution of packages from the public npm registry at runtime if they are not already installed or cached in the environment.
[PROMPT_INJECTION]: The skill ingests untrusted data in the form of filenames and configuration file contents (such as package.json or pyproject.toml) and uses them to determine execution logic. While standard for developer tools, this represents an indirect injection surface if the project files contain instructions designed to manipulate the agent's behavior during the detection phase.

format