The Agent Skills Directory

[SAFE]: The skill implements a comprehensive research orchestration system with deterministic scripts for idea ranking, source mapping, and execution feasibility analysis.
[COMMAND_EXECUTION]: The skill uses subprocess.run to execute standard research tools such as git, nvidia-smi, and rocm-smi to gather environment data and manage isolated experiment branches. These calls are scoped to local environment discovery and project maintenance.
[EXTERNAL_DOWNLOADS]: Remote data fetching is limited to well-known academic and developer services (arXiv, DOI/Crossref, and GitHub APIs) for resolving research metadata. These operations use standard Python libraries (urllib.request) with restrictive timeouts.
[DYNAMIC_EXECUTION]: The system uses importlib to perform controlled syntax and constructor checks on local repository files during the smoke-testing phase. These probes are gated by safety checks that exclude entry-point scripts like train.py or eval.py to prevent unintentional full-scale execution during the planning phase.

ai-research-explore