The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from local files and remote URLs to perform uploads, creating a surface for potentially malicious instructions embedded in that data to influence agent behavior.
Ingestion points: The file_path and file_url parameters in the upload_attachment_to_card and upload_attachment_to_table_record tools (SKILL.md, references/mcp.md).
Boundary markers: Absent; the instructions do not define delimiters or provide specific warnings to the agent regarding the content of the processed files.
Capability inventory: Tools for uploading attachments to Pipefy cards and table records, and for creating S3 presigned upload URLs (SKILL.md).
Sanitization: An SSRF guard is documented for the file_url parameter, restricting downloads to public HTTPS hosts (references/mcp.md).- [DATA_EXFILTRATION]: The skill facilitates a read-and-send chain where content from the local filesystem or external URLs is transmitted to the vendor's infrastructure. While Pipefy is the intended destination, this capability provides a mechanism to access and transmit sensitive local data if abused.- [COMMAND_EXECUTION]: The skill documentation includes examples and instructions for using the Pipefy CLI (pipefy attachment upload) to perform file operations, which involves the execution of shell commands in the user's environment.- [EXTERNAL_DOWNLOADS]: Through the file_url parameter, the skill can be instructed to perform network requests to external domains to retrieve file content for uploading.

pipefy-attachments