Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/ig.pyexecutes system commands to process media content. - Evidence: The script uses
subprocess.run()to callyt-dlp,ffprobe, andffmpegto fetch and transcode video content. - Context: The implementation follows security best practices by passing arguments as a list rather than a shell string, effectively preventing shell command injection through the user-provided URL.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external Instagram URLs that could contain malicious instructions designed to influence the agent's behavior during analysis.
- Ingestion points: External metadata such as
descriptionandtitleare fetched from Instagram URLs inscripts/ig.pyviayt-dlp. - Boundary markers: The instructions in
SKILL.mddo not define delimiters or provide specific warnings to the agent to ignore embedded instructions in the processed metadata. - Capability inventory: The skill has the capability to write files to the local filesystem (
~/clawd/media/) and execute media processing binaries. - Sanitization: Metadata strings are truncated to 500 characters, but no content filtering or escaping is applied to prevent the agent from interpreting the text as commands.
Audit Metadata