The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external Zeek dns.log files, which ingest untrusted network data. This creates a surface where an attacker could embed malicious instructions within DNS queries to influence the agent's behavior when it interprets the analysis results.
Ingestion points: The scripts/agent.py script reads and parses log files provided via the --log-file argument.
Boundary markers: The skill lacks explicit boundary markers or instructions to the agent to treat data from the logs as untrusted or to ignore embedded instructions.
Capability inventory: The skill can read local files and write analysis reports to the filesystem (via the --output argument) but lacks network or system command execution capabilities.
Sanitization: No sanitization or escaping is performed on the DNS query strings extracted from the logs before they are included in the final report.
[METADATA_POISONING]: There is a discrepancy in the author attribution between files; the SKILL.md frontmatter lists 'mahipal' as the author, whereas the LICENSE file attributes copyright to 'mukul975'. Inconsistent metadata can be used to obscure the true origin or ownership of the skill content.

detecting-exfiltration-over-dns-with-zeek