The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides explicit instructions and commands for extracting highly sensitive credentials from network traffic. Examples include extracting HTTP POST data (http.file_data), session cookies (http.cookie), HTTP authorization headers (http.authorization), and FTP passwords (ftp.request.arg). This enables full harvesting of cleartext credentials passing through monitored interfaces.
[COMMAND_EXECUTION]: The skill relies on the execution of the tshark binary and several local shell scripts, including capture-http-credentials.sh, analyze-dns-queries.sh, and extract-files-from-capture.sh. It encourages the use of an -x flag to execute operations rather than just displaying them.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests and processes untrusted data from network interfaces and packet captures.
Ingestion points: Live network interfaces (eth0, en0) and local pcap files.
Boundary markers: None identified in the provided instructions; however, the skill mentions a PreToolUse hook that validates targets against a .pentest/scope.json file.
Capability inventory: Execution of tshark (network capture/analysis) and bash (script execution).
Sanitization: No specific sanitization or escaping of captured network payloads is mentioned before being presented to the agent.
[DYNAMIC_CONTEXT_INJECTION]: The skill utilizes the ! command syntax in SKILL.md to execute shell commands at load time. These commands are used for environment discovery, specifically to verify if tshark is installed and to check for the presence of specific wrapper scripts in the scripts/tshark/ directory.

tshark