The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/monitor-server.sh script is vulnerable to command injection. In functions like test_model_connectivity and test_provider, user-supplied provider names are taken from JSON request bodies and interpolated into curl commands using double quotes (e.g., "' + test_url + '"). Because double quotes allow command substitution in shell environments, an attacker can execute arbitrary code using the $(...) or backtick syntax.
[CREDENTIALS_UNSAFE]: Multiple scripts and documentation examples (including scripts/create-provider.sh, scripts/set-model.sh, and the README.md) demonstrate passing sensitive AI provider API tokens as command-line arguments. Secrets passed this way are visible in system process listings (ps) and shell history, posing a risk of credential exposure.
[COMMAND_EXECUTION]: The monitor-server.sh component uses subprocess.run with shell=True and string concatenation to build commands. This pattern is inherently dangerous as it relies on perfect sanitization of all inputs, which is not present in this skill's implementation.
[CREDENTIALS_UNSAFE]: The scripts/install.sh script interpolates sensitive environment variables, such as HICLAW_MANAGER_GATEWAY_KEY, directly into JSON strings used in shell-executed commands. This can lead to credential leakage or command injection if the key contains special shell characters.

ai-gateway-management