fal-realtime
Pass
Audited by Gen Agent Trust Hub on Apr 3, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/realtime.shexecutes shell commands includingcurlfor network requests andjqfor parsing JSON responses. - [DATA_EXFILTRATION]: The skill transmits user-provided image prompts and the
FAL_KEYto the fal.ai API athttps://fal.run. This is the intended behavior for the skill. - [SAFE]: The script implements a mechanism to store the
FAL_KEYin a local.envfile, which aligns with standard environment variable management practices. - [SAFE]: The skill contains a surface for indirect prompt injection due to its handling of user input.
- Ingestion points: User input is ingested via the
--promptargument inscripts/realtime.sh. - Boundary markers: No markers are present to isolate user input from the JSON structure.
- Capability inventory: The skill performs external network operations using
curl. - Sanitization: User input is directly interpolated into the JSON payload string without escaping or validation, which may lead to malformed API requests if the prompt contains control characters like double quotes.
Audit Metadata