The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes instructions to install necessary analysis tools (Ghidra, binutils, QEMU, GDB) using official system package managers like apt, dnf, pacman, and Homebrew. These are considered safe, well-known sources for developer tools.
[INDIRECT_PROMPT_INJECTION]: The skill processes external binary files, which creates an attack surface for indirect prompt injection if the binaries contain instructions the agent might interpret or if they exploit vulnerabilities in the analysis tools.
Ingestion points: Instructions guide the agent to load and analyze target.bin and target.elf files as specified in the '架构确认' and '入口定位' sections.
Boundary markers: There are no explicit delimiters used to separate the external data from the agent's instructions within the skill body.
Capability inventory: The skill utilizes several powerful utilities including file, readelf, objdump, qemu-riscv64/32, and gdb-multiarch to interact with and execute parts of the binary data.
Sanitization: The skill mitigates this risk by referencing a 'default sandbox + network isolation' policy for any dynamic execution of the target binaries, referring to platform-specific safety guidelines.

re-riscv