The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill is designed to execute arbitrary Bash commands provided via the command argument. It uses the pattern sleep <seconds> && <command> without any validation or sanitization, allowing for the execution of any system command.
[REMOTE_CODE_EXECUTION] (HIGH): The direct execution of unvalidated user/LLM input in a shell environment constitutes a remote code execution vulnerability if the agent is manipulated into running malicious payloads.
[DATA_EXFILTRATION] (MEDIUM): Arbitrary command execution can be leveraged to read sensitive local files (e.g., ~/.ssh/id_rsa) and exfiltrate them via network commands like curl or wget.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill presents a vulnerability surface for indirect prompt injection. 1. Ingestion points: The command argument in SKILL.md. 2. Boundary markers: Absent. The input is interpolated directly into a shell command string. 3. Capability inventory: Full Bash shell execution including subprocess spawning and background tasks. 4. Sanitization: Absent. No filtering or validation is performed on the input string before execution.

delayed-command