filmcraft-product-video

Fail

Audited by Gen Agent Trust Hub on Oct 23, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The file references/craft-command-evidence.json specifies a remote download for a portable Windows executable: https://github.com/storytold/filmcraft/releases/download/v0.8.0/filmcraft-0.8.0-windows-x64-portable.zip. This binary is used as the primary tool for the skill's operations.
  • [COMMAND_EXECUTION]: The skill provides detailed instructions for executing the downloaded filmcraft-cli.exe through the shell. The execution pattern includes passing project files and generated command lists: filmcraft-cli --project input.fcproj --save-as output.fcproj run steps.jsonl.
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted media assets from the user.
  • Ingestion points: Media files are imported using the file.import method with user-provided paths.
  • Boundary markers: There are no explicit instructions or delimiters defined to separate user-provided content from the command logic.
  • Capability inventory: The agent has the capability to perform file system operations (read/write) and execute shell commands via the CLI tool.
  • Sanitization: No sanitization or validation logic is present to ensure that imported media or its metadata does not influence the command execution flow.
  • [DYNAMIC_EXECUTION]: The skill instructions involve creating a JSONL file (steps.jsonl) at runtime which contains a sequence of commands to be processed and executed by the filmcraft-cli tool. This is a form of dynamic code generation and execution.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 23, 2026, 10:15 AM