The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to process untrusted binary data from malware samples (e.g., beacon.bin, sample.bin, memory.dmp). If these files contain strings specifically crafted to look like instructions, the agent might interpret them as authoritative commands when they are extracted and displayed in the context.
Ingestion points: The skill reads external files using BeaconConfig.from_path("beacon.bin") in SKILL.md and open(sys.argv[1], "rb").read() in references/cs-beacon-decrypt.md.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore or treat extracted content as untrusted data.
Capability inventory: The skill utilizes yara (SKILL.md), python3 (SKILL.md, references/cs-beacon-decrypt.md), vol3 (SKILL.md), and strings (SKILL.md, references/modern-c2-analysis.md) for data processing and analysis.
Sanitization: String values are decoded using .decode(errors='replace'), which prevents binary parsing errors but does not filter out malicious natural language instructions.
[COMMAND_EXECUTION]: The skill provides numerous commands for the agent to execute shell-based analysis tools.
Evidence: Instructions to run yara -r c2_rules.yar sample.bin, vol3 -f memory.dmp windows.memmap, and various Python analysis scripts like 1768.py and parse_beacon_config.py in SKILL.md and the reference files.

c2-beacon-analysis