The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves external data (memories and knowledge base documents) via the MemOS API and incorporates it into the agent's context, creating a potential surface for indirect prompt injection.
Ingestion points: Memory content retrieved from search_memory and get_user_profile in scripts/memos_cloud/operations.py.
Boundary markers: The SKILL.md instructions include a directive to "use only relevant memories; ignore noise" to mitigate the impact of irrelevant or malicious retrieved data.
Capability inventory: The skill possesses network capabilities (HTTPS POST via scripts/memos_cloud/client.py) and local file-reading capabilities (via scripts/memos_cloud/files.py for KB uploads).
Sanitization: No specific sanitization or filtering of retrieved memory text is performed before it is provided to the agent.
[COMMAND_EXECUTION]: The skill uses a Python-based CLI wrapper (scripts/memos_cloud.py) to execute its logic. The execution is limited to pre-defined operations using the argparse library in scripts/memos_cloud/cli.py, which helps prevent arbitrary command injection.
[SAFE]: The skill communicates exclusively with the official vendor domain memos.memtensor.cn. Authentication is handled securely via environment variables (MEMOS_API_KEY), and the documentation explicitly warns against exposing these secrets to the LLM.

memos-cloud-server