clips-studio

Pass

Audited by Gen Agent Trust Hub on Aug 27, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests user-supplied text to generate video content and audio via an external provider (fal.ai). A lack of technical sanitization on these inputs creates a surface for indirect prompt injection, where malicious instructions could attempt to bypass provider safety filters or the skill's own 'honesty' guidelines.\n
  • Ingestion points: User-provided prompts and motion briefs in SKILL.md stages 1, 2, and 3.\n
  • Boundary markers: Prompts are delimited by quotes when passed to the CLI script in SKILL.md, but there is no validation of the content itself.\n
  • Capability inventory: The scripts/falvid.py tool performs network egress to api.fal.ai and writes binary video files to the local file system.\n
  • Sanitization: No input sanitization or content filtering is implemented in scripts/falvid.py.\n- [CREDENTIALS_UNSAFE]: The scripts/falvid.py script contains a utility function _autoload_fal_keys that automatically searches the current working directory and the user's home directory (~) for files that might contain API credentials. The script identifies targets based on keywords (e.g., 'fal', 'key', 'api', '.env') and reads their full content to search for the FAL_KEY pattern. This behavior involves automated discovery and reading of potentially sensitive local files to extract secrets stored in plain text.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 27, 2026, 02:06 AM
Security Audit — agent-trust-hub — clips-studio