sue-synccode

What problem does it solve?

Manually syncing experiment code to remote sandboxes is error-prone, risks overwriting valuable run artifacts, and wastes time debugging SSH access or git authentication issues. This Skill eliminates those risks by enforcing strict path contracts and automated sync workflows.

Core Features & Use Cases

• Strict Path Governance: Enforces DeepResearch root and workspace identity rules to prevent path mismatches that could break experiment runs or corrupt outputs.
• Dual Sync Modes: Supports both git and rsync/scp transfer methods, with automatic fallback to rsync/scp if sandbox GitHub SSH authentication fails, so sync never blocks on key debugging.
• Inode-Safe Exclusions: Automatically excludes virtual environments, caches, and runtime artifacts from transfers to save sandbox inodes and preserve existing experiment outputs.
• Use Case: When preparing to run a scale-up ML experiment on LUMI or RunPod, use this Skill to push your latest code changes to the sandbox without accidentally overwriting previous checkpoint or result files.

Quick Start

Use the sue-synccode skill to sync your latest DeepResearch experiment code to the selected remote sandbox before running dryrun or fullrun validation.