The Agent Skills Directory

[SAFE]: The skill operates locally as a data processing tool, using pandas and tldextract to analyze proxy and DNS logs. All operations are confined to parsing user-provided files and generating reports, with no evidence of command execution or network exfiltration.
[PROMPT_INJECTION]: The skill processes untrusted data from network logs, which presents a surface for indirect prompt injection. However, the analysis is restricted to structured data extraction (e.g., domain names, byte counts), which effectively sanitizes the input and prevents the execution of embedded instructions. 1. Ingestion points: parse_proxy_log, parse_dns_log, and parse_csv_log in scripts/agent.py. 2. Boundary markers: The output is structured JSON, though no explicit 'ignore instructions' markers are used. 3. Capability inventory: Local file read access for log processing. 4. Sanitization: Regex-based field extraction and type conversion (e.g., integer casting for bytes) are used.
[SAFE]: There is a minor metadata discrepancy where the author is listed as 'mahipal' in SKILL.md but copyright is attributed to 'mukul975' in the LICENSE file; however, this is consistent with the provided author context and does not indicate malicious poisoning.

detecting-shadow-it-cloud-usage