The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external JSON data that directly influences filesystem operations. The as_of field in the input JSON is used to generate filenames without sanitization, which could lead to path traversal vulnerabilities.
Ingestion points: scripts/build_review_queue.py reads user-provided JSON files as its primary data source.
Boundary markers: The script lacks explicit boundaries or safety instructions to separate data content from processing logic.
Capability inventory: The script writes report files to the filesystem using the pathlib library.
Sanitization: There is no validation to prevent path traversal characters (e.g., ../) in the as_of field used to construct output paths.
[COMMAND_EXECUTION]: The skill's test suite utilizes the subprocess module to run the main script during automated checks.
Evidence: scripts/tests/test_build_review_queue.py uses subprocess.run() to execute scripts/build_review_queue.py. Although this is restricted to internal testing and uses fixed arguments, it represents a command execution attack surface.

kanchi-dividend-review-monitor