Skip to main content

velum-labs/handoffkit

SkillsMP has collected 6 skills from velum-labs/handoffkit. Open a skill to review its source and details.

Latest recorded source activity
SkillsMP catalog refreshed
skills collected
6
GitHub stars
1
GitHub forks
0

Skills in this repository

2 occupation categories · 100% classified

Showing 6 of 6 collected skills.

occupation
Data Scientists
description

Supervise a hyperkit hypergrid search that hill-climbs FusionKit's open-weight compound toward closed-frontier SOTA on a locked holdout. Use when asked to "run the hypergrid", "supervise the hill climb", "find the best open-weight compound", "close the gap to…

updated
occupation
Software Developers
description

Operate the shared HyperKit experiment lab in lab/ on behalf of one operator: propose a new experiment, lock and submit a sweep, monitor/extend it, or conclude it with results. Use when asked to "propose an experiment", "add an experiment to the lab", "claim…

updated
occupation
Software Developers
description

Answer questions about the shared HyperKit experiment lab in lab/ without changing anything: what experiments exist, what is running, what the latest results and decisions were, who owns what, which follow-ups are up for grabs, and whether a proposed idea…

updated
occupation
Software Developers
description

Coordinates cross-repo releases for the velum-labs workspace (handoffkit, cursorkit, fusionkit, mlx-lm) using a Terraform-style plan/apply flow. Use when asked to "cut a release", "release", "ship velum", "publish packages", "plan a release", "apply a…

updated
occupation
Software Developers
description

Self-healing hill climb that drives FusionKit's fused compound to provably beat the best single panel model on an unsaturated benchmark. Use when asked to "hill climb fusion", "improve the fusion benchmark", "make the compound beat the best model", "reduce…

updated
occupation
Software Developers
description

Autonomous end-to-end production audit of FusionKit's fusion value: set up the environment, run billed public benchmarks (OpenAI + Anthropic panel), save all artifacts, score docs/fusion/FUSION_VALUE_RUBRIC.md, hill climb until the fused compound provably…

updated
Showing 6 of 6 collected skills.