The Agent Skills Directory

Data Exposure & Exfiltration (MEDIUM): The skill explicitly targets ~/Library/Logs/SkillLauncher/app.log. These logs contain a history of commands executed by the user (发送: /skill-name 参数), which may include sensitive arguments, tokens, or personal information processed by other skills.
Indirect Prompt Injection (MEDIUM): This skill creates an attack surface for indirect prompt injection by reading untrusted data (application logs).
Ingestion points: ~/Library/Logs/SkillLauncher/app.log via tail and grep commands.
Boundary markers: Absent. The agent is instructed to "analyze" the logs without specific delimiters to distinguish between log metadata and potentially malicious log content.
Capability inventory: The skill uses shell commands (tail, grep, date) which implies the agent has command execution permissions in the underlying environment.
Sanitization: None. The agent processes raw log lines which could contain adversarial instructions injected into the logs via previous commands.
Command Execution (LOW): The skill provides bash snippets for log rotation and searching. While using standard tools like grep and tail, it encourages the agent to execute shell commands on files that contain arbitrary user-controlled strings.

log-viewer