The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as network logs, email content, and web request parameters to identify threats. While the logic is analytical, this data ingestion creates a surface for indirect prompt injection. 1. Ingestion points: Methods within scripts/detection_utils.py including detect_malware_behavior and detect_phishing accept event data from external sources. 2. Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore commands within the data. 3. Capability inventory: The skill can write files to a user-specified directory via DetectionRuleSet.export_all. 4. Sanitization: Data is analyzed via regex and keyword matching, but not transformed to prevent agent obedience.
[SAFE]: The reputation scanner alert for SKILL.md is considered a false positive. The skill documentation includes examples of malware behavior, ransomware patterns, and exploit payloads for the purpose of teaching detection rule creation. These strings match common malware signatures but are used here in a defensive context.
[EXTERNAL_DOWNLOADS]: The skill references established security resources including the MITRE ATT&CK framework and SIGMA rules repository. These references are used for documentation and rule mapping, posing no risk to the execution environment.

detection