Skill Claude CodeCodex
Delegate long-running, sequential ML experiments to one economical high-reasoning subagent while the primary agent retains scientific decisions and independently verifies frozen inputs, completed artifacts, gates, and verdicts. Use for multi-seed GPU campaigns, benchmark matrices, confirmation runs, locked…
Skill Claude CodeCodex
Design, implement, and execute reproducible cross-modal representation-loss studies on real data. Use when comparing an incumbent contrastive objective with kernel/HSIC/CKA/KTA, CCA, VICReg, Barlow-style, or multi-positive alternatives; building deterministic GPU pilot and confirmation campaigns; auditing duplicate…
Skill Claude CodeCodex
Run the vibescience-mcp scientific-experiment-log loop end to end — pair deep research (perplexityresearch) + arxiv MCP with a real, verdict-computed experiment cycle, then dogfood it into the markdown vault. Use when testing research hypotheses on real data and logging predicted-vs-observed causal claims for…
MCP server Claude CodeCodexCursor
MCP server "vibescience-mcp" as configured in AntonIXO/vibescience-mcp. Runs locally from the vibescience-mcp Python package.