The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains multiple functional examples for executing arbitrary code across different runtimes.
references/games-and-vms-3.md includes explicit instructions for crafting ROP chains using mprotect, read, and shellcode to escape C-based jails, which triggered a shellcode-generation antivirus alert.
references/pyjails.md documents eval, exec, and compile bypasses for Python environments, including decorator-based injection to reach builtins from restricted scopes.
references/games-and-vms-3.md describes marshal code injection vectors to execute arbitrary Python code objects on a server.
[PRIVILEGE_ESCALATION]: Extensive methodology for acquiring elevated permissions in Linux environments.
references/linux-privesc.md details exploitation of sudo wildcard parameter injection via fnmatch, SUID binary abuse in cronjobs, and the Polkit integer overflow (CVE-2018-19788).
Documentation covers LD_PRELOAD hijacking of allowed binaries in restricted shells (rbash).
Bypasses for restricted vim (rvim) using python3 execution or netrw file browsing are provided.
[COMMAND_EXECUTION]: Provides patterns for spawning shells and executing system commands.
references/bashjails.md includes reverse shell patterns such as cat <&3 | bash >&3 2>&3 using /dev/tcp virtual sockets.
Documents techniques for spawning interactive shells from character-restricted environments using $0 expansion (\$$#).
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from CTF platforms and challenge files without explicit sanitization.
Ingestion points: Challenge artifacts including jails, encodings, and multi-stage puzzles identified in SKILL.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the provided templates.
Capability inventory: Includes toolsets for network operations (netcat), signal processing (aircrack-ng), and Python scripting which could be misused if an agent obeys instructions embedded in a challenge.
Sanitization: Lacks sanitization logic for data retrieved from CTFd APIs or local challenge files.
[OBFUSCATION]: Comprehensive coverage of techniques used to hide malicious intent or bypass filters.
Documentation includes octal encoding in ANSI-C quoting ($'...'), Unicode homoglyph bypasses for Python identifiers, and multi-layer encoding chains (Base64 inside Hex inside URL-encoding).
Describes exotic encodings like base65536 (CJK Unicode) and April Fools' RFC formats like UTF-9.
[DATA_EXFILTRATION]: Documents methods for sending data to external attacker-controlled servers.
references/bashjails.md describes using /dev/tcp for raw TCP exfiltration when standard tools like curl or wget are missing.
references/dns.md details data exfiltration via DNS tunneling and encoded subdomains.
[PERSISTENCE]: Includes techniques for maintaining access across sessions.
references/bashjails.md mentions modifying shell profiles such as ~/.bashrc to maintain access.

misc-ctf