The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it is designed to read and process project records, assets, and Structured Text which could contain malicious instructions embedded by external contributors.
Ingestion points: Data enters the agent context through client.items.find, client.uploads.list, and collection traversal via listPagedIterator (as described in references/records.md and references/filtering-and-pagination.md).
Boundary markers: The instructions lack specific guidance or delimiters to ensure the agent ignores natural language instructions found within the project content during transformation or backfill tasks.
Capability inventory: The skill maintains high-privilege capabilities including CLI command execution (datocms-cli), network operations via the CMA client (client.request), and the ability to generate local code deliverables.
Sanitization: There is no explicit requirement for the sanitization or escaping of record content before it is interpolated into agent workflows or scripts.
[SAFE]: No malicious patterns, obfuscation, or unauthorized data exfiltration were detected. The skill actively promotes security by instructing the agent to read credentials from environment variables and to never hardcode secrets.
[EXTERNAL_DOWNLOADS]: The skill references official NPM packages provided by the vendor (e.g., @datocms/cma-client-node, datocms-structured-text-utils) to facilitate API interactions and document transformations. These are standard resources for the intended skill usage.

datocms-cma