The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to analyze external source code and architectural metadata, which are untrusted ingestion points.
Ingestion points: The skill is designed to process user-provided Kotlin source code, Android project structures, and existing analysis documents (e.g., references/auri-analysis.md).
Boundary markers: None identified. The skill assumes the validity of the structure it analyzes without explicit delimiters for untrusted content.
Capability inventory: The skill utilizes provided scripts (complexity_analyzer.py, dependency_graph.py) that can read file contents and write analysis reports to the local filesystem.
Sanitization: The skill performs no explicit sanitization or filtering of the code it processes beyond regex-based metric extraction.
[COMMAND_EXECUTION]: The skill documentation encourages and instructs the agent to execute specific Python scripts included in the distribution.
Evidence: SKILL.md contains explicit instructions like "run: python complexity_analyzer.py C:/project" and "run: python dependency_graph.py C:/project".
The scripts perform automated analysis by walking directory trees and reading file contents.
[DATA_EXFILTRATION]: The scripts provided in the scripts/ directory lack comprehensive path validation for input directories.
Evidence: In scripts/complexity_analyzer.py and scripts/dependency_graph.py, the safe_user_path function is implemented to restrict file writing to the current working directory. However, this check is only applied to the --output argument. The primary path argument (the directory to be analyzed) is used to initialize a Path object without validation, allowing the scripts to traverse and read metadata/code from arbitrary filesystem paths outside the workspace.

matematico-tao