The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to extract and display content from potentially untrusted files (e.g., binary strings, HTTP request headers, and image metadata). This content is presented directly to the agent's context without sanitization or explicit instructions to disregard embedded commands.
Ingestion points: pcap_extract_http.py (HTTP payloads/DNS queries), steg_quickcheck.py (Metadata/Strings), and SKILL.md (LSB extraction snippets).
Boundary markers: The output is preceded by headers like [HTTP Request] or [*] Stage 3: Strings extraction, but lacks strong delimiters to prevent the agent from following instructions hidden within the extracted data.
Capability inventory: The skill possesses the capability to execute shell commands (subprocess.run), write to the filesystem (/tmp/steg_extract.bin), and perform network analysis.
Sanitization: Minimal; content is primarily decoded as latin-1 or UTF-8 with errors ignored and sometimes truncated to 500 characters, which does not prevent instructional content from being processed by the LLM.
[COMMAND_EXECUTION]: The skill utilizes subprocess.run and shell scripts to execute external forensic tools. While the Python scripts pass arguments as lists (minimizing shell injection risks), the skill's core functionality relies on executing a variety of local system commands based on user-provided file paths.
Evidence: network_forensics/pcap_extract_http.py executes tshark and tcpflow. steganography/steg_quickcheck.py executes exiftool, strings, steghide, zsteg, and stegoveritas.

forensics-tools