The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes telemetry and behavioral reports from external and internal analysis tools. Extracted indicators of compromise (IOCs) such as IPs and domains are pushed to firewall blocklists without manual review or validation.
Ingestion points: Malware analysis reports in scripts/agent.py and SKILL.md retrieved from sandbox services.
Boundary markers: Uses structured JSON for API interactions to separate data from commands, though it does not validate the semantic content of indicators.
Capability inventory: Capabilities include reading quarantined files (MalwareCollector) and issuing network requests to modify security policies (push_iocs_to_blocklist).
Sanitization: No validation or reputation checking is applied to extracted values before pushing them to blocking infrastructure, creating a potential for service disruption.
[DATA_EXFILTRATION]: The skill performs network operations to internal security services and external APIs. It includes a mechanism to bypass TLS certificate validation via the SKIP_TLS_VERIFY environment variable. While documented for development/lab use, this configuration reduces the security of data transmitted to security infrastructure.

building-automated-malware-submission-pipeline