The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The documentation explicitly lists default credentials (admin / agentiko) for a management web dashboard running at http://localhost:9119.
[PROMPT_INJECTION]: The skill describes a "YOLO mode" feature that intentionally bypasses safety filters and user approval prompts for dangerous commands like file deletion (rm) or offensive probes, allowing the agent to act without user confirmation.
[COMMAND_EXECUTION]: The skill provides instructions for executing arbitrary commands on a remote worker via SSH and includes a specific workaround to bypass platform restrictions (where write_file might be blocked) by using Python's os.chmod() to create and execute shell scripts.
[REMOTE_CODE_EXECUTION]: The cron job implementation allows for the persistent and automated execution of arbitrary instructions and toolsets, which can include loading additional external skills into automated sessions.
[DATA_EXFILTRATION]: The skill provides a specialized Python script designed to bypass WAF protections (specifically Vercel/Next.js Security Checkpoints) by extracting React Server Components (RSC) content from raw HTML payloads.

agentiko-hermes