The Agent Skills Directory

[PROMPT_INJECTION]: No attempts to override safety guidelines or agent behavior were found in the instructions.
[DATA_EXFILTRATION]: No sensitive file access, credential exposure, or network communication patterns were detected. The mentioned websites (e.g., TechCrunch, Reddit) are used as generic examples.
[REMOTE_CODE_EXECUTION]: The skill does not download or execute any external scripts or packages.
[COMMAND_EXECUTION]: No shell commands, privilege escalation attempts, or persistence mechanisms are present.
[OBFUSCATION]: No hidden content, encoded strings, or homoglyphs were identified.
[INDIRECT_PROMPT_INJECTION]: While the skill guides the agent to process external data (news, blogs, forums), it lacks the capabilities (like file system writes or command execution) to be exploited via indirect injection.

research-sources