The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill manages MongoDB credentials and documentation indicates that passwords can be provided via the --password command-line flag. This practice exposes sensitive information in the process environment and shell history. Furthermore, the skill documentation mentions a fallback to plaintext configuration storage if native OS secret stores are unavailable.
[COMMAND_EXECUTION]: The skill facilitates the execution of the agent-mongo CLI binary through the Bash tool to interact with MongoDB instances.
[DATA_EXFILTRATION]: The skill handles MongoDB connection strings and credentials which are transmitted to external database servers during connection and query operations.
[PROMPT_INJECTION]: The skill ingests untrusted data from external MongoDB databases, creating a surface for indirect prompt injection.
Ingestion points: Data is ingested through the output of query find, query get, query aggregate, and collection schema as documented in SKILL.md and references/commands.md.
Boundary markers: The CLI outputs data in structured NDJSON or JSON formats, which provides clear boundaries between individual records and metadata.
Capability inventory: The skill uses Bash(agent-mongo *), Read, Grep, and Glob tools for data processing and system interaction.
Sanitization: The tool converts BSON types to JSON strings and implements a default truncation of string fields exceeding 200 characters, which limits the volume of data processed in a single line.

agent-mongo