The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a workflow for processing untrusted data which could potentially lead to indirect prompt injection attacks if not properly handled.
Ingestion points: External data is ingested through the fcode.context.parameters object in the entry point main() function.
Boundary markers: The documentation does not suggest the use of specific delimiters or markers to separate untrusted parameter data from instruction context.
Capability inventory: The skill provides access to several sensitive capabilities, including triggering other processes (fcode.processes.run), uploading files to storage (fcode.storage.upload), and sending emails (fcode.send_mail).
Sanitization: There is a lack of explicit guidance or examples showing how to sanitize or validate these input parameters before they are used as arguments for the platform's powerful toolset, which could allow malicious input to influence the agent's behavior or platform state.

fcode-python