The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to process user-defined models.py and resources.yaml files to automate the generation of database client code and CloudFormation resources. This ingestion of untrusted local data represents an attack surface where maliciously crafted project files could influence the agent's output. * Ingestion points: models.py (containing model definitions) and resources.yaml (containing application configuration). * Boundary markers: No explicit delimiters or instructions are provided to the agent to separate the ORM's logic from potentially malicious user-provided definitions. * Capability inventory: The toolchain generates Python source files (prismarine_client.py), modifies resource definitions in the SAM template, and configures DynamoDB stream triggers for Lambda execution. * Sanitization: The instructions do not describe any sanitization or validation mechanisms for the ingested metadata.
[DYNAMIC_EXECUTION]: The integration workflow for EasySAM (Stage 1) involves the dynamic importing of local Python files from user-defined packages to extract model metadata and validate schemas. * Execution method: Python import operations are performed on project files during the easysam generate or easysam deploy lifecycle. * Scope: The inspection process executes the code within models.py to register model classes via the Cluster object.

prismarine-skill