The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and execute a script from a remote URL directly into a shell sub-process.
Evidence: bash -c "$(curl -fsSL https://gef.blah.cat/sh)" found in SKILL.md.
[PRIVILEGE_ESCALATION]: The skill uses sudo to acquire administrative permissions for package installation and system configuration.
Evidence: sudo apt-get install -y gdb and echo 0 | sudo tee /proc/sys/kernel/randomize_va_space found in SKILL.md and references/malware-debugging.md.
[EXTERNAL_DOWNLOADS]: The skill downloads external repositories and scripts to the local system from sources that are not categorized as trusted vendors.
Evidence: git clone https://github.com/pwndbg/pwndbg and curl -fsSL https://gef.blah.cat/sh.
[DYNAMIC_EXECUTION]: The skill utilizes GDB's Python API to define and execute scripts that dynamically interact with process memory, registers, and execution flow.
Evidence: Python scripts in SKILL.md and references/scripting.md using gdb.execute(), frame.read_register(), and inf.write_memory().
[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is analyzing untrusted binary data (malware), creating a surface where malicious input could influence agent behavior.
Ingestion points: gdb -q ./binary and read_memory() calls in SKILL.md and references/scripting.md.
Boundary markers: None present.
Capability inventory: Shell access, memory manipulation, file system writes, and network tracing.
Sanitization: None present.

gdb