skills/npc-live/clawfirm/cover-gen/Gen Agent Trust Hub

cover-gen

Warn

Audited by Gen Agent Trust Hub on May 20, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The provided workflow includes shell scripts that use eval $(ffprobe ...) to parse video file metadata. This pattern is vulnerable to command injection if a video file contains malicious strings in its metadata fields (such as title or comment tags), which the shell would then execute during evaluation.\n- [PROMPT_INJECTION]: The skill defines a prompt template in SKILL.md for generating covers that interpolates untrusted user data ({title}, {desc}) without using boundary markers or sanitization, creating a surface for indirect prompt injection.\n
  • Ingestion points: SKILL.md (Step 3c: prompt template; Step 4: platform script variables)\n
  • Boundary markers: Absent\n
  • Capability inventory: File system access (/tmp/, ~/Desktop/), subprocess execution (ffmpeg, ffprobe, media-gen binary)\n
  • Sanitization: No evidence of sanitization for user-provided variables like $TITLE or $VIDEO_INPUT before being used in shell commands or AI prompts.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 20, 2026, 06:07 AM
Security Audit — agent-trust-hub — cover-gen