The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute multiple Python scripts using shell commands (e.g., bash -c "python3.11 scripts/detect.py ."). These scripts (detect.py, verify-citation.py, render-pdf.py, etc.) are not provided within the skill package, making their behavior unverifiable.
[REMOTE_CODE_EXECUTION]: The skill references an external implementation repository at https://github.com/ipunithgowda/firewall-review. The IMPLEMENTATION_STATUS.md file notes that this repository was inaccessible (404) during validation. Relying on an external, untrusted, and potentially missing repository for core functionality is a significant security risk.
[EXTERNAL_DOWNLOADS]: The skill references information from several official sources for compliance standards, including nist.gov, pcisecuritystandards.org, iso.org, and cisecurity.org. These are well-known and trusted organizations.
[INDIRECT_PROMPT_INJECTION]: 1. Ingestion points: The skill processes firewall configuration files located in the Pre-requisites/ directory. 2. Boundary markers: No explicit boundary markers or "ignore" instructions were found for the configuration processing. 3. Capability inventory: The skill has the ability to execute shell commands, write files to the Deliverables/ folder, and uses LLM-based sub-agents for analysis. 4. Sanitization: While a redaction script is mentioned for the learning loop, its code is not included in the skill. The skill uses LLM sub-agents (like senior-pentester) to analyze raw configuration data. Maliciously crafted content within these configuration files could influence the agent's behavior, leading to incorrect audit findings or other unintended consequences.

firewall-review