research-dev-standards
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to perform environment and resource management using standard system utilities. This includes disk space checks (
lsblk,df -hT), GPU monitoring (nvidia-smi), and container orchestration (docker pull,docker build,docker run). It also specifies rendering workflows usingnodeandpython3for Blender and Three.js tasks. - [EXTERNAL_DOWNLOADS]: The skill facilitates the retrieval of model weights, datasets, and configuration assets from well-known research repositories and archives, including GitHub (e.g.,
mrdoob/three,matplotlib/cmocean) and arXiv. These operations are conducted with a focus on security, requiring SHA256 verification and the use of atomic file naming conventions. - [INDIRECT_PROMPT_INJECTION]: A potential risk surface exists as the skill processes external datasets and third-party model weights. While the skill mandates integrity checks, it does not explicitly define sanitization protocols for the content of these external files.
- Ingestion points: Upstream dataset downloads and model weight imports defined in
references/dataset-storage.md. - Boundary markers: Absent; no specific instructions are provided to the model to ignore potential instructions embedded in processed data.
- Capability inventory: Shell access (
bash), container control (docker), and documentation automation via Notion workflows. - Sanitization: Absent; the workflow relies on source integrity rather than content filtering.
Audit Metadata