The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute arbitrary shell commands provided by the user or agent via arguments.
Evidence: In watch.sh, the command is executed using eval "$COMMAND" (line 228).
Evidence: In watch.js, the command is executed using execSync(cmd, ...) (line 79).
[INDIRECT_PROMPT_INJECTION]: The skill processes file system data (filenames and paths) and interpolates them into the execution environment, creating a vulnerability surface for command injection.
Ingestion points: File system event data (filenames and paths) in watch.sh (line 330) and watch.js (line 118).
Boundary markers: None present. The variables are passed directly into the environment.
Capability inventory: Arbitrary shell command execution via eval (Bash) and execSync (Node.js).
Sanitization: No sanitization or shell-escaping is performed on file paths before they are made available to the shell environment. A file named with shell metacharacters (e.g., ; rm -rf /) could trigger unintended command execution if the provided command string references the $FILE variable unsafely.

file-watcher