claude-gif-optimize

Fail

Audited by Gen Agent Trust Hub on Oct 7, 2026

Risk Level: HIGHPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill's error handling instructions suggest using sudo apt install gifsicle. This pattern encourages the agent to execute commands with root privileges, which is a significant security risk as it allows for persistent system modification beyond the skill's intended scope.
  • [COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to execute system binaries and internal scripts located in the user's home directory.
  • Evidence: Execution of ~/.video-skill/bin/python3 ~/.claude/skills/claude-gif/scripts/gif_optimize.py.
  • Evidence: Use of complex command substitution in Strategy 6: $(seq -f '#%g' 0 2 $(gifsicle --info INPUT.gif | grep -oP '\d+ images' | grep -oP '\d+')).
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits a vulnerability surface where untrusted data from GIF files is processed and used to influence future agent actions.
  • Ingestion points: File metadata from INPUT.gif read in Step 2 and Step 4 (Strategy 6).
  • Boundary markers: Absent; there are no explicit delimiters or instructions to ignore embedded content in the processed metadata.
  • Capability inventory: Bash (command execution) and Read (file system access).
  • Sanitization: Limited; although grep is used to filter for digits in one instance, there is no comprehensive sanitization of tool output before it is passed to shell commands.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 7, 2026, 07:48 AM