The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill provides troubleshooting instructions that involve system-level security modifications.
Evidence: In references/troubleshooting.md, users are instructed to run sudo ufw disable to resolve WebRTC streaming issues, which reduces the host system's security posture.
[DYNAMIC_EXECUTION]: The skill utilizes a component designed to execute custom code at runtime.
Evidence: The udfloader element, described in SKILL.md and references/api-and-pipelines.md, allows the microservice to load and execute Python User Defined Functions (UDFs) within the GStreamer pipeline.
[INDIRECT_PROMPT_INJECTION]: The skill handles external data ingestion and has write capabilities, creating an attack surface for malicious data to influence the system.
Ingestion points: The agent ingests external video sources and streams via the source.uri parameter in REST API requests (defined in SKILL.md).
Boundary markers: No explicit boundary markers or safety instructions are defined for the handling of external media content.
Capability inventory: The skill can execute shell commands via docker compose, perform network requests via curl, and write data to the local file system using the gvametapublish element (documented in references/api-and-pipelines.md).
Sanitization: The instructions do not specify sanitization or validation procedures for the ingested URI sources or resulting metadata.

dlsps-user