The Agent Skills Directory

[DATA_EXFILTRATION]: The skill performs inspection of sensitive process memory and layout information.
Evidence: The script scripts/vdso_gadget_extract.py reads /proc/self/mem, /proc/self/maps, and /proc/self/auxv to identify the base address of the vDSO and scan for executable code gadgets within the process's own memory space.
[COMMAND_EXECUTION]: The skill includes a custom Python script and instructions for using various system diagnostic and compilation tools.
Evidence: scripts/vdso_gadget_extract.py is provided for identifying syscall gadgets; instructions in references/elf-format.md suggest the use of readelf, objdump, gcc, and gdb for inspecting and debugging target binaries.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to parse and interpret complex external data sources, creating a potential attack surface if the inputs are malicious.
Ingestion points: The agent parses ELF binary headers, section metadata, and process information from the /proc filesystem as described in references/elf-format.md and references/procfs.md.
Boundary markers: Absent. The skill does not provide delimiters or specific instructions to ignore instructions that might be embedded in binary metadata, section names, or process strings.
Capability inventory: The skill includes memory inspection via a Python script and utilizes external utilities like gdb and readelf for data extraction.
Sanitization: Absent. There is no evidence of validation or filtering for data ingested from external binaries or the system environment before it is processed by the agent.

linux-internals-dev