The Agent Skills Directory

Indirect Prompt Injection Surface: The skill manages user-provided design context by reading from a persistent local file, user-context.md. As this data influences the agent's future tasks, it creates a potential surface for indirect prompt injection. This behavior is fundamental to the skill's purpose and is tempered by clear safety guidelines.
Ingestion points: The scripts/user_context_preflight.py script is used to read and process the user-context.md file located in the state directory.
Boundary markers: The processing script parses the markdown file into a structured JSON format, which helps delineate the stored data from the agent's operational instructions.
Capability inventory: The scripts associated with this skill are restricted to local file operations and do not include network access or dynamic code evaluation capabilities.
Sanitization: The skill relies on clear instructional boundaries and specific warnings to the agent rather than programmatic sanitization of the stored text.
Local Command Execution: The skill utilizes local Python scripts for initialization and state management. These scripts are self-contained, using standard libraries to perform routine file I/O operations without requiring elevated privileges or external dependencies.
Data Persistence: The skill maintains context across sessions by storing information in a local directory. It follows security best practices by explicitly instructing the agent not to save sensitive information like credentials, secrets, or API keys.

user-context