The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted binary files, which creates an inherent attack surface for indirect prompt injection if the agent interprets data extracted from those binaries as instructions.
Ingestion points: The scripts/string_triage.py script reads local binary artifacts provided by the user. The skill also suggests using various third-party static analysis tools (e.g., readelf, binwalk, Ghidra) to inspect these artifacts.
Boundary markers: The instructions explicitly warn the agent to default to static analysis and not to execute unknown binaries or contact embedded endpoints unless explicitly authorized in an isolated lab environment.
Capability inventory: The skill utilizes file system access for reading artifacts, hashing, and metadata extraction. It instructs the agent to use external disassemblers and decompilers to build behavior maps of target binaries.
Sanitization: The internal triage script escapes common control characters (like newlines and carriage returns) when formatting extracted strings for display to the agent, reducing the risk of accidental instruction execution from the artifact's data.
[SAFE]: The skill's code and instructions focus on defensive security analysis and static triage. No patterns of data exfiltration, credential harvesting, unauthorized remote code downloads, or persistence mechanisms were detected. The included triage script uses only standard Python libraries and does not execute the target binary.

binary-reversing