The Agent Skills Directory

[PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface because its analysis scripts (scripts/architecture_diagram_generator.py, scripts/dependency_analyzer.py, and scripts/project_architect.py) ingest untrusted data from the user's local project directory. Specifically, directory names, file names, and class names are extracted and presented to the AI agent in the form of architectural reports and diagrams.
Ingestion points: The ProjectScanner and PatternDetector classes in the Python scripts read the file system structure and file contents of the provided project_path to identify components and layers.
Boundary markers: Although the output is formatted as structured Mermaid, ASCII, or JSON, there are no explicit boundary markers or instructions telling the agent to ignore potentially malicious instructional content embedded within the project's structural metadata.
Capability inventory: The skill has the ability to read the project structure and write result files to the local disk using the --output command-line argument, which could be leveraged if an injection succeeds.
Sanitization: The scripts do not sanitize or filter the content of directory or file names for instructional language before including them in the final report provided to the agent.

senior-architect