The Agent Skills Directory

COMMAND_EXECUTION (CRITICAL): The script scripts/parse_crash.py is vulnerable to shell command injection. When the -c (content) flag is used, the script joins the raw, unsanitized crash content into a command string and executes it via os.system(). An attacker can inject arbitrary commands (e.g., using ;, &&, or |) that will be executed with the privileges of the agent.
REMOTE_CODE_EXECUTION (HIGH): The vulnerability in parse_crash.py facilitates remote code execution if the agent is instructed to analyze a crash log obtained from an untrusted remote source, such as a URL or a third-party application log.
PROMPT_INJECTION (MEDIUM): This skill is highly vulnerable to indirect prompt injection. Malicious instructions or payloads can be embedded in crash data, which the agent then passes to the vulnerable tool, resulting in system compromise without the user's direct involvement.
DATA_EXFILTRATION (LOW): The script scripts/screenshot.py utilizes hdc file recv to automate the transfer of files from a connected device to the host. While intended for screenshots, this capability could be used to extract sensitive files from a HarmonyOS device if the agent is manipulated by a malicious prompt.

ohos-app-build-debug