The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/feasibility_profile.py performs runtime compilation by generating gcc commands with dynamic flags to build a 'witness' binary from source. It then executes the resulting binary using safe_subprocess to empirically verify vulnerability triggers.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a high attack surface for indirect injection as it ingests untrusted external data including vulnerability_list, crash_corpus, and compiled target binaries. These ingestion points are paired with high-privilege capabilities such as arbitrary subprocess execution and network communication, without explicit input sanitization or boundary markers.
[COMMAND_EXECUTION]: Multiple scripts, including scripts/exploit_context.py, scripts/offset_finder.py, and scripts/rop_autochain.py, execute external CLI tools and user-provided binaries (e.g., pwn checksec, patchelf, and the exploitation targets) using subprocess calls.
[REMOTE_CODE_EXECUTION]: While intended for security research, the skill automates the generation and delivery of exploit payloads (ROP chains, heap corruption) to remote network services via pwntools.remote(). The execution of these payloads represents the primary function of the skill but carries significant risk if used against unintended or malicious targets.

exploit-development