The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a dedicated section instructing the agent on how to bypass the 'Tirith' security scanner, which is designed to protect system and credential files. It provides specific workarounds, such as using Python open() and shell heredocs, to evade tool-level path restrictions.
[COMMAND_EXECUTION]: The environment operates with root privileges, giving the agent full control over the container. It facilitates the use of powerful networking tools (socat, nmap, masscan) and provides instructions for setting up persistent task execution via the cronjob utility.
[REMOTE_CODE_EXECUTION]: The skill encourages the creation and execution of arbitrary shell and Python scripts in /root/scripts/. It guides the agent on using execute_code combined with os.chmod() to dynamically generate and run executable files.
[EXTERNAL_DOWNLOADS]: The instructions direct the agent to use the apk package manager to install additional software and runtimes (Go, Node.js) from external repositories during the session.
[CREDENTIALS_UNSAFE]: The documentation references the location of sensitive SSH private keys (/opt/data/ssh/agentiko_key and /root/.ssh/) used for container-to-container communication, making them accessible to the agent.
[DATA_EXFILTRATION]: The included reference file (rsc-extraction.md) provides methodologies for bypassing external security checkpoints and web application firewalls (e.g., Vercel Security Checkpoint) to extract data from remote targets that attempt to block automated access.

agentiko-worker