The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions explicitly direct the agent to circumvent security restrictions imposed by the 'Tirith security scanner'. It provides multiple techniques (Python open(), heredocs, and Base64 chunking via terminal) to bypass write_file blocks on protected paths such as /root/output/ and /root/scripts/. These instructions include directing subagents to also utilize these bypass methods.
[COMMAND_EXECUTION]: The skill facilitates the execution of arbitrary shell commands across a cluster of containers (worker, worker-heavy, worker-tor) via SSH. It provides access to a powerful toolset including gcc, make, gdb, strace, and various network scanners, with explicit instructions for compiling exploits and PoCs on the worker-heavy instance.
[DATA_EXFILTRATION]: The environment utilizes a shared SSH private key located at /opt/data/ssh/worker_key for lateral movement between containers. The presence of worker-tor configured for anonymous network operations provides a mechanism for covert data exfiltration or probing hostile infrastructure while bypassing geographical or rate-limiting controls.
[EXTERNAL_DOWNLOADS]: The skill includes a dedicated reference for bypassing WAFs and Security Checkpoints (specifically Vercel/Next.js) by extracting React Server Components (RSC) payloads. This involves downloading and processing potentially untrusted external content with the intent of evading security filters.

worker