higgsfield-facs
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides instructions for executing a local validation script via
python3 scripts/seedance_lint.py. This script accepts a user-influenced prompt as a command-line argument, representing a potential surface for argument injection if the execution environment does not adequately sanitize shell characters. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user data, such as character descriptions and monologues, to generate structured prompts for video and image models. While it uses specific block markers, there is a risk of instruction leakage from user-supplied text into the final generated prompt.
- Ingestion points: User-supplied monologues in dialogue blocks and character descriptions for image/video generation in
SKILL.md. - Boundary markers: The skill uses structured tags like
[AUDIO: Xs]and numbered beat schedules (Beat 1 (0-1s): ...) to separate components. - Capability inventory: The skill executes local shell scripts (
scripts/seedance_lint.py) and interfaces with external image and video generation tools. - Sanitization: No specific mention of escaping or filtering logic for user-provided dialogue before it is placed into audio or performance blocks.
- [DYNAMIC_EXECUTION]: The workflow involves dynamically constructing and running a shell command (
python3 scripts/seedance_lint.py --preflight --model seedance_2_0 "<prompt>") where the<prompt>variable is derived from user content, which is a common dynamic execution pattern used for validation.
Audit Metadata