The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the configuration of an external MCP server at https://rube.app/mcp. This endpoint is used to provide the Coda automation tools and acts as the interface between the agent and the Coda API.
[DATA_EXFILTRATION]: The skill possesses tools to read and extract sensitive information from Coda, including document content (CODA_GET_A_PAGE), table data (CODA_LIST_TABLE_ROWS), and sharing metadata (CODA_GET_SHARING_METADATA). Users should be aware that this data is processed through the external Rube MCP infrastructure.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from external sources and has powerful write/administrative capabilities.
Ingestion points: Document content and table data are retrieved via CODA_GET_A_PAGE, CODA_LIST_TABLE_ROWS, and CODA_SEARCH_ROW.
Boundary markers: The instructions do not define boundary markers or delimiters to help the agent distinguish between administrative instructions and document data.
Capability inventory: The skill includes high-impact tools such as CODA_UPSERT_ROWS (modify data), CODA_ADD_PERMISSION (modify access control), and CODA_PUBLISH_DOC (make documents public).
Sanitization: There is no mention of sanitization or validation of the content retrieved from Coda before it is processed by the agent.

coda-automation