The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the execution of local Python scripts (search.py, core.py, design_system.py) to perform search operations and generate design documentation on the host system.
[DATA_EXFILTRATION]: A directory traversal vulnerability exists in scripts/design_system.py within the persist_design_system function. The script uses the user-supplied project_name to construct a file path without sanitizing for traversal sequences like .. or /. This allows an attacker (or a misled AI agent) to create directories and files (specifically MASTER.md) in arbitrary locations on the file system where the user has write permissions.
[PROMPT_INJECTION]: The skill exhibits a vulnerability to Indirect Prompt Injection. It ingests data from multiple CSV files and interpolates the content directly into generated markdown files (MASTER.md and page-specific overrides). The AI agent is explicitly instructed to treat these generated files as a 'Master' source of truth and follow their rules, creating a channel where malicious content embedded in the source data could override agent behavior.
Ingestion points: The query and project_name arguments in scripts/search.py, as well as the content of the CSV files in the data/ directory.
Boundary markers: No delimiters or safety warnings are present in the generated markdown files to prevent the agent from obeying instructions found within the data.
Capability inventory: The skill has the capability to write to the file system using Python's open() function with 'w' mode.
Sanitization: While the BM25 search tokenizer removes punctuation, the data extracted from CSVs is used raw in the document generation process.

dioxus-ui-ux