The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves content from external, user-populated sources such as Knowledge Bases and shared memory, incorporating this data into the agent's context window.
Ingestion points: Data is ingested through the memory_search and knowledgebase_search tools as described in SKILL.md and references/tools-rag-workflow.md.
Boundary markers: While the operating contracts recommend limiting retrieval to the top 10 items and abstaining if evidence is missing, there are no explicit instructions to use structural delimiters (e.g., XML tags) to isolate retrieved content from agent instructions.
Capability inventory: The agent has the ability to write to long-term memory (memory_add), upload files (artifact_put), and modify conversation history (conversation_add_message), which are the primary targets for manipulation via indirect injection.
Sanitization: The skill does not mandate specific sanitization or filtering of content retrieved from the Knowledge Base before it is processed.
[DATA_EXFILTRATION]: The skill is designed to transmit conversation history, user preferences, and file artifacts to the Meko Cloud infrastructure for persistence.
Evidence: Tools such as artifact_put, memory_add, and conversation_add_message facilitate the transfer of data to the vendor's domains (mekodata.ai), as outlined in references/tools-cookbook.md.
Context: This behavior is the intended primary function of the skill, providing durable agent memory and collaborative knowledge management. The data transfers are directed to the vendor's legitimate services.

meko-mcp-tools