The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests external data in the form of a JSON encryption schema or manual user input to drive its calculations. This untrusted input serves as a potential surface for indirect prompt injection where instructions could be embedded within the data values.
Ingestion points: User-provided encryption schema (pasted content or file path) requested in Step 2.
Boundary markers: The instructions do not define specific delimiters or "ignore instructions" wrappers for the external schema content.
Capability inventory: The skill has the capability to write calculation results to a temporary file (qe-sizing-calculations.md) on the host system.
Sanitization: The skill includes a dedicated "Validation" section that restricts inputs to specific BSON types and valid query configurations, which serves to mitigate the risk of processing malicious data.
[DYNAMIC_EXECUTION]: The skill instructs the agent to perform mathematical operations based on formulas provided in the instructions (e.g., storage and memory impact formulas). This involves the runtime evaluation of expressions containing user-provided variables like average byte length (v), lower bounds (lb), and upper bounds (ub). While this is a form of dynamic execution, the risk is reduced because the inputs are validated as integers and the formulas themselves are static templates provided by the skill author.
[SAFE]: The skill implements a proactive security measure by explicitly instructing the agent to never ask for or accept sample document data, preventing the accidental exposure of sensitive plaintext information during the estimation process.
[SAFE]: The skill utilizes the system's temporary directory for calculation logs and includes a final step that reminds the user to delete the generated file to protect metadata regarding the database schema and field lengths.

mongodb-qe-size-estimation