The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs the agent to download and install various third-party security utilities from external repositories.
References tools like GEF (github.com/bata24/gef), pwndbg (github.com/pwndbg/pwndbg), and libc-database (github.com/niklasb/libc-database).
Instructs usage of package managers (apt, pip, gem) to install dependencies like pwntools, ROPgadget, ropper, and one_gadget.
[COMMAND_EXECUTION]: Provides numerous shell commands for environment setup, binary forensics, and exploit orchestration.
Includes commands for binary inspection (checksec, readelf, strings), package installation, and kernel environment manipulation (cpio, qemu-system-x86_64).
[REMOTE_CODE_EXECUTION]: The primary function of the skill is to generate remote code execution (RCE) exploit scripts.
Contains Python/pwntools templates and C code snippets designed to achieve shell access on targets via stack overflows, heap corruption, and kernel vulnerabilities.
[PRIVILEGE_ESCALATION]: Detailed documentation and code are provided for escalating privileges, specifically within the Linux kernel.
Covers techniques such as ROP chains to call commit_creds(prepare_kernel_cred(0)) and hijacking modprobe_path to achieve root access.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect injection as it instructs the agent to ingest and analyze untrusted binary files and memory leaks.
Ingestion points: Binary files (./vuln), libraries (./libc.so.6), and kernel interface data (/proc/kallsyms).
Boundary markers: Absent in the logic that processes strings or symbol data from the analyzed files.
Capability inventory: Full access to shell execution, network sockets (remote()), and local file system operations.
Sanitization: No specific sanitization logic is provided for handling strings extracted from untrusted binary targets.

pwn-chain