BloomAgent
BloomAgent contient 7 skills collectées depuis jonastbrg, avec une couverture métier par dépôt et des pages de détail sur le site.
Skills dans ce dépôt
Analyze and visualize BLOOM evaluation results across probes, models, and conditions. Use: /bloom-analyze [--probe <name>] [--model <name>]
Run a full BLOOM behavioral evaluation pipeline for a probe. Orchestrates all 4 stages. Use: /bloom-eval <probe_name>
Stage 2 only: generate diverse trigger scenarios for a probe. Use: /bloom-ideate <probe_name> [--n N]
Stage 4 only: score all traces for a probe and write judgment.json files. Use: /bloom-judge <probe_name>
Stage 3 only: run all trials for a probe (spawns parallel subagents). Use: /bloom-rollout <probe_name>
Run BLOOM evaluations at scale across multiple probes, models, and repetitions. Use: /bloom-sweep --probe <name> --models <m1,m2> --scenarios N --reps N [--setup] [--parallel]
Stage 1 only: deeply analyze a behavioral probe and write understanding.json. Use: /bloom-understand <probe_name>