The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted binary samples, which represents a surface for indirect prompt injection attacks.
Ingestion points: The scripts/quick_triage.sh script and the strings utility ingest data from an external, untrusted binary file provided as an argument.
Boundary markers: The triage script uses fenced section headers (e.g., === IDENTITY ===, === STRING CLUSTERS ===) to delimit different types of analysis output.
Capability inventory: The skill possesses capabilities to execute shell commands (e.g., file, binwalk, readelf), run local Python scripts, and perform file system reads to generate reports.
Sanitization: The skill employs strings -a -n 6 to extract printable characters and utilizes head to truncate bulk output, which provides a layer of data filtering but does not fully prevent malicious strings from appearing in the output.
[COMMAND_EXECUTION]: The skill facilitates the execution of various local analysis tools and bundled scripts.
Evidence: scripts/quick_triage.sh invokes tools like binwalk, readelf, and nm, and executes the included entropy.py and string_clusters.py Python scripts to process the user-provided sample file.

reverse-malware-triage