The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted external JSON data which is then searchable by an AI agent. This creates a surface for indirect prompt injection attacks if the search results contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Data is loaded into the search engine via engine.Load(stream) in C# projects (references/csharp.md) and through the POST /api/teams/{team}/datasets/{dataset}/load endpoint in the HTTP API (references/http-api.md).
Boundary markers: The documentation does not provide specific guidance on using boundary markers or delimiters to isolate search results from agent instructions in the prompt.
Capability inventory: The AI agent is provided with tools to list available datasets, describe field capabilities, perform full-text searches, and retrieve full document records (SKILL.md).
Sanitization: No explicit content sanitization or prompt injection filtering is described for the data being indexed or retrieved.
[SAFE]: No hardcoded credentials or unauthorized data exfiltration patterns were detected. Sensitive configuration values, such as API tokens and JWT signing keys, are handled using placeholders and standard secret management practices (e.g., environment variables and dotnet user-secrets).
[SAFE]: External dependencies, including the IndxSearchLib NuGet package, NPM packages under the @indxsearch scope, and the IndxCloudApi GitHub repository, belong to the verified vendor. Other referenced tools, such as axios and mcp-remote, are well-known services or ecosystem tools.
[SAFE]: The skill does not contain any obfuscated code, hidden instructions, or suspicious persistence mechanisms. Operations involving remote code (e.g., git clone or npx) target official vendor repositories or public package registries.

indx-search