The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies extensively on executing the genmedia command-line utility for various tasks including model discovery, schema inspection, asset management, and model execution. This involves passing user-supplied arguments directly to the shell.
[DATA_EXFILTRATION]: The genmedia upload <local_file_or_url> command allows the agent to read arbitrary local files and upload them to the fal.ai infrastructure. This presents a data exposure risk if the agent is manipulated into uploading sensitive local files such as credentials, private keys, or configuration files (e.g., ~/.ssh/id_rsa, .env).
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data by passing task descriptions and prompts directly to external tools and APIs (genmedia models "<task>", genmedia run ... --prompt "<value>"). This creates a vulnerability surface for indirect prompt injection.
Ingestion points: User-provided task strings in genmedia models and parameter values (like prompts) in genmedia run and assets characters create.
Boundary markers: The skill does not define explicit delimiters or instructions to ignore embedded commands within the processed data.
Capability inventory: The skill possesses capabilities for command execution (genmedia CLI), file reading (via upload), and file writing (via run --download).
Sanitization: There is no mention of sanitization or validation of the input strings before they are interpolated into the CLI commands.

genmedia