The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation instructions in .claude/skills/awf-skill/README.md and SKILL.md include commands to download and execute a script from the developer's official GitHub repository via a pipe to bash. This is the primary installation method for the tool.
[PRIVILEGE_ESCALATION]: Several tools provided by the skill use sudo to perform administrative tasks, such as configuring network namespaces, managing iptables rules, and interacting with the Docker daemon. These permissions are necessary for the firewall's core functionality.
[OBFUSCATION]: The one-shot-token library used by the firewall employs XOR encoding to store protected environment variable names in its binary. This is a documented defensive measure designed to prevent the extraction of these names via static analysis tools like strings.
[INDIRECT_PROMPT_INJECTION]: The awf-debug-tools skill includes Python scripts that parse Squid proxy access logs. Since these logs contain domain names and other metadata from external network requests, they represent a potential surface for indirect prompt injection if an attacker can influence the logged traffic.
Ingestion points: .claude/skills/awf-debug-tools/scripts/parse-squid-logs.py
Boundary markers: Absent in the raw log file structure.
Capability inventory: The debug scripts use subprocess to execute Docker commands and inspect the host environment.
Sanitization: No explicit sanitization or filtering of the log content is performed before processing.
[COMMAND_EXECUTION]: The debug tools make extensive use of subprocess and docker exec to inspect container health, analyze logs, and verify network connectivity.

awf-skill