The Agent Skills Directory

Prompt Injection (HIGH): The skill contains explicit behavioral overrides in SKILL.md that instruct the agent to disregard its intended cron tool and instead use the generic exec or bash tools. This steers the agent away from constrained environments toward high-privilege shell access.
Command Execution (HIGH): By mandating the use of the exec tool to manage cron jobs (e.g., exec: clawdbot cron add ...), the skill creates a significant command injection surface. If an attacker can influence the contents of the cron name or message, they may be able to execute arbitrary shell commands on the host system.
Indirect Prompt Injection (MEDIUM): The skill teaches the agent to use a "robust message template" that embeds natural language instructions ([INSTRUCTION: DO NOT USE ANY TOOLS]) within data fields. This pattern acknowledges a lack of boundary enforcement and demonstrates how untrusted data can be used to steer agent behavior at runtime.
External Downloads (LOW): The README.md suggests installation via clawdhub, an unverified third-party repository that is not part of the trusted source list.

ez-cronjob