The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides templates for using curl to interact with the Fabric API. These templates use quoted heredocs (e.g., <<'JSON'), which is a best practice for passing multi-line data to commands without the risk of shell interpolation or command injection.
[EXTERNAL_DOWNLOADS]: The skill communicates with the official Fabric API endpoint at api.fabric.so. This is a legitimate service associated with the skill's primary purpose and is treated as a safe source.
[CREDENTIALS_UNSAFE]: Authentication is handled via the FABRIC_API_KEY environment variable. The skill correctly instructs users to provide this sensitive token through their environment configuration rather than hardcoding it in scripts.
[PROMPT_INJECTION]: The skill processes user-supplied data for fields such as notepad text and folder names. This represents a potential surface for indirect prompt injection, but it is handled securely at the execution layer.
Ingestion points: Untrusted data enters the agent context via user-provided values for fields like name, text, url, and description in SKILL.md and the OpenAPI schema.
Boundary markers: Templates in SKILL.md use quoted heredocs (<<'JSON'), which prevent the shell from expanding variables or executing commands embedded in the data.
Capability inventory: The skill utilizes curl to perform REST API operations (GET, POST) on the Fabric platform.
Sanitization: Content is passed directly into JSON structures within the heredoc; while no NLP-level filtering is performed, the shell-level transport is properly escaped.

fabric-api