The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes natural language input and database data (schema definitions, query logs, document samples), creating a surface for indirect prompt injection.
Ingestion points: natural language user input (SKILL.md, Access Pattern Analysis section), database content via MCP tools (SKILL.md, MongoDB MCP Integration section).
Boundary markers: The skill includes an explicit 'Action Policy' section requiring user confirmation for write/destructive operations.
Capability inventory: MCP tools provide capabilities for reading and writing to MongoDB collections (find, aggregate, update-many, insert-many).
Sanitization: The skill relies on manual user review and approval for any database modifications; automated sanitization is not specified.
[EXTERNAL_DOWNLOADS]: The skill references official MongoDB repositories on GitHub for the MCP server and source code imports.
Evidence: References to github.com/mongodb-js/mongodb-mcp-server and github.com/mongodb/agent-skills were identified.
Context: These are well-known, trusted sources associated with the vendor and do not present a security risk.
[COMMAND_EXECUTION]: The skill instructs the agent to execute MongoDB shell commands and aggregation pipelines for diagnostic purposes.
Evidence: Examples include db.serverStatus().opcounters and administrative queries like db.getSiblingDB("admin").aggregate([{ $queryStats: {} }]) found in references/source-query-stats.md.
Context: These commands are standard for database performance analysis and are gated by the skill's safety policy requiring user approval for non-read operations.

mongodb-schema-design