The Agent Skills Directory

COMMAND_EXECUTION (HIGH): The scripts/scan-all.sh script is vulnerable to shell command injection via the $DIR parameter. It uses eval to process a condition string built from this unsanitized input. An attacker can execute arbitrary commands by providing a directory name containing shell metacharacters (e.g., '; touch /tmp/pwned; #').
CREDENTIALS_UNSAFE (HIGH): The scripts/scan.sh script prints the raw, unredacted lines containing detected secrets directly to standard output. While the SKILL.md specifies a mandatory redaction format (showing only the first and last 4 characters), the implementation uses grep or ripgrep to output the full secret, leading to sensitive data exposure in the agent's logs.
COMMAND_EXECUTION (MEDIUM): scripts/scan-all.sh attempts to execute scripts from hardcoded paths in the user's home directory (~/.config/opencode/skill/). This creates a dependency on external files and could lead to the execution of untrusted code if an attacker can manipulate files in those locations.

security-secrets