The Agent Skills Directory

[COMMAND_EXECUTION]: The documentation includes a troubleshooting command ssh root@172.245.56.50 "crontab -l". This instruction encourages users to connect to a hardcoded external IP address with root privileges, which is a high-risk practice that could facilitate unauthorized access or system compromise.
[PROMPT_INJECTION]: The skill exhibits a vulnerability surface for indirect prompt injection through its automated data ingestion process.
Ingestion points: Data is fetched from external platforms Kie.ai and fal.ai to update the model observatory database via daily-sync.ts and kie-observatory.worker.ts.
Boundary markers: The skill body does not specify any boundary markers or instructions to disregard potential malicious directives embedded within the fetched model metadata.
Capability inventory: The ingested data populates ai_model_recommendations, which is used by the agent to make autonomous decisions about which AI models to use for complex pipeline tasks.
Sanitization: There is no indication that the content retrieved from external sources is sanitized or validated before being integrated into the agent's decision-making context.

model-observatory