The Agent Skills Directory

[METADATA_POISONING]: Discrepancy between the described purpose of the skill's script and its actual implementation. The scripts/README.md and the internal documentation of scripts/analyze_request_patterns.py state that the script identifies serial network requests for parallelization. However, the provided code only performs basic file system statistics, such as counting files, measuring total size, and identifying file extensions.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze application source code to identify request patterns, which introduces a potential attack surface where instructions hidden in the processed data could influence agent behavior.
Ingestion points: Application files and file metadata accessed via the Read tool and the directory traversal logic in scripts/analyze_request_patterns.py.
Boundary markers: Absent. There are no instructions or delimiters defined to prevent the agent from executing instructions found within the analyzed application files.
Capability inventory: The skill is granted Bash capabilities (including curl, ping, and traceroute) and Write access, which provide potential vectors for impact if an injection occurs.
Sanitization: Absent. The skill does not implement validation or sanitization of the content it extracts from the target application.

analyzing-network-latency