The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes a technique to bypass application-level validation by nesting restricted database operations within allowed stages. This represents an attack surface where an agent might be used to generate or execute malicious injection payloads.
Ingestion points: User-controlled pipeline parameter in API requests (SKILL.md).
Boundary markers: The skill specifically details how to evade existing server-side stage allowlists.
Capability inventory: The skill utilizes curl and Python script execution to interact with remote endpoints.
Sanitization: It highlights the lack of recursive sanitization for nested MongoDB aggregation stages.
[DATA_EXFILTRATION]: The primary objective of the documented technique is the unauthorized extraction of sensitive information from a database.
The skill identifies targets such as "invite tokens, password hashes, session docs, API keys" from non-public collections.
[COMMAND_EXECUTION]: The skill recommends using command-line tools and scripts to automate the exploitation process.
Evidence includes instructions for using curl with --data-urlencode and Python helpers to deliver injection payloads.

api-mongo-agg-facet-bypass