The Agent Skills Directory

[PROMPT_INJECTION]: The skill documentation describes a workflow that is inherently susceptible to indirect prompt injection.
Ingestion points: The system is designed to ingest external content through the add() method and API endpoints (/v3/documents), which process text, URLs, and multi-modal files (references/sdk-guide.md).
Boundary markers: Integration examples demonstrate the use of XML-style delimiters like <context> and <user_profile> to separate retrieved memory from system instructions (references/sdk-integrations.md).
Capability inventory: The agent is explicitly instructed to use retrieved memories to personalize its behavior and generate contextually aware responses (references/use-cases.md).
Sanitization: There is no mention of specific sanitization, filtering, or validation steps for ingested external data before it is presented to the LLM.
[EXTERNAL_DOWNLOADS]: The skill references several official and well-known software packages.
Python packages: Mentions official vendor packages supermemory, supermemory-openai-sdk, and supermemory-pipecat, as well as well-known libraries such as langchain-openai, crewai, and the OpenAI agents SDK (references/sdk-guide.md, references/sdk-integrations.md).
Node.js packages: Mentions official vendor packages supermemory and @supermemory/tools, alongside trusted community packages like @ai-sdk/openai and langchain (references/sdk-guide.md, references/sdk-integrations.md).
GitHub Repositories: Provides links to the official supermemoryai GitHub organization for open-source resources (README.md).

supermemory