The Agent Skills Directory

[COMMAND_EXECUTION]: The server-side code in server/server.py and server/runpod_handler.py utilizes subprocess.run() with shell=True to execute ffmpeg for video optimization. The input filenames for these operations are generated from video titles extracted via yt-dlp from social media URLs. If a video title contains shell metacharacters, it could lead to command injection in the host environment.
[EXTERNAL_DOWNLOADS]: During initialization, the skill downloads brain atlas files (.annot) from the Thomas Yeo Lab's public GitHub repository. It also installs the tribev2 dependency directly from Meta's research repository on GitHub. These are recognized sources for the skill's stated neuroscience purpose.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from social media URLs, user-provided text, and media files. This data is ingested into the agent's context for neuro-marketing analysis, which represents an attack surface for instructions embedded within the processed media or text. Evidence chain:
Ingestion points: The social_url, text, and video_url parameters in the /analyze API endpoint within server/server.py.
Boundary markers: The SKILL.md file defines report structures but does not include explicit delimiters or warnings to ignore command-like text within the analyzed content.
Capability inventory: The skill has the capability to execute shell commands (subprocess.run), write to the file system (tempfile), and make network requests (requests, urllib).
Sanitization: There is no evidence of specific sanitization or filtering for the content before it is processed by the model or analyzed by the agent.
[DATA_EXPOSURE]: Deployment scripts (scripts/deploy_to_persistent.sh, scripts/download_and_analyze.py) include a hardcoded IP address (13.221.72.26) used for testing or demonstration. Furthermore, the AWS provisioning script (scripts/launch_persistent.sh) configures security groups to allow unrestricted ingress traffic from all IPs (0.0.0.0/0) on ports 22 and 8000, creating a significant network security risk for the deployed infrastructure.

meta-tribe-skill