The Agent Skills Directory

[OBFUSCATION]: The skill provides detailed instructions for implementing reverse-engineering resistance and hiding intent from security analysts.
Evidence includes recommendations in references/offensive-lowlevel.md for using obfstr and litcrypt to perform compile-time XOR encryption of strings such as URLs, mutexes, and file paths to hide Indicators of Compromise (IOCs).
It suggests using garble (a rustc-driver obfuscator) for control-flow and identifier renaming.
It recommends a release profile that strips all symbols and aborts on panic to minimize the diagnostic information available to analysts.
[DYNAMIC_EXECUTION]: The skill describes several methods for executing code dynamically, including patterns that bypass traditional filesystem monitoring.
references/dynamic-dispatch-and-plugins.md details "in-memory loading" via manual mapping on Windows, which involves manual allocation of memory, copying PE sections, and resolving import tables without using standard Windows loaders.
It describes file-less loading on Linux using memfd_create to create an anonymous file descriptor for an ELF binary and loading it via /proc/self/fd/.
[PERSISTENCE]: The skill explicitly includes persistence as a core module in its recommended architecture for offensive tooling.
In references/offensive-lowlevel.md, the "Offensive architecture patterns" section describes a modular monolith where "persistence" is listed alongside "encrypt, spread, and exfil" as standard capabilities.
[COMMAND_EXECUTION]: The skill encourages the use of direct syscalls to bypass operating system security APIs.
references/offensive-lowlevel.md details "Hell's Gate" family patterns for resolving syscall numbers at runtime by walking the PEB/export table and issuing syscalls via inline assembly to avoid detection by security software monitoring standard library calls.
[PRIVILEGE_ESCALATION]: The skill describes methods for manipulating memory protections that are commonly used in exploit development and shellcode execution.
references/offensive-lowlevel.md provides patterns for flipping page permissions (e.g., RW to RX) using VirtualProtect on Windows or mprotect on Linux to execute code in memory regions that were previously non-executable.
[EXTERNAL_DOWNLOADS]: The skill suggests mechanisms for binaries to download and execute code from remote sources.
references/dynamic-dispatch-and-plugins.md recommends self_update for pulling releases from GitHub, GitLab, and other platforms to automatically swap the running binary with a remote version.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a broad attack surface as it is designed to process and review arbitrary Rust source code.
Ingestion points: The skill ingests untrusted code during PR reviews or refactoring tasks as specified in SKILL.md.
Boundary markers: There are no instructions provided to the agent to treat code blocks as untrusted data or to ignore embedded instructions within processed files.
Capability inventory: The skill has access to extensive high-risk capabilities including FFI boundaries, unsafe blocks, subprocess execution (Command::new), and network operations for self-updating.
Sanitization: There are no described mechanisms for sanitizing or escaping the content of Rust files before they are processed by the agent.

rust-patterns