The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests metadata and sample values from a remote Typesense instance to generate instructions for the AI agent, creating a surface for indirect prompt injection.
Ingestion points: The scripts/generate_schema_context.js script fetches collection schemas and samples facet values via the client.collections().retrieve() and documents().search() methods.
Boundary markers: The prompt generated by the script does not use specific delimiters or instructions to treat database metadata as untrusted content.
Capability inventory: The skill scripts can perform network requests to the configured Typesense host and can write files to the local filesystem using the --output flag in generate_schema_context.js.
Sanitization: The script performs no escaping or sanitization of field names, collection names, or enum values before including them in the final LLM prompt output.
[COMMAND_EXECUTION]: The skill relies on the execution of local scripts and system commands to interact with the Typesense service.
Evidence: SKILL.md and README.md provide instructions for executing npm install, node scripts (e.g., generate_schema_context.js, search.js, create_collection.js), and curl commands for document management.

typesense