wandb-autoresearch — Claude Code skill: what it does, how to install
wandb-autoresearch runs bounded machine learning trials via W&B Launch, smoke-tests real jobs, compares metrics, and tracks research state across sessions.
Apache-2.0 387 installs ★ 76 updated 2026-10-23
npx skills add wandb/skills --skill wandb-autoresearch -a claude-codeWhat wandb-autoresearch does
The skill coordinates training research through W&B Launch instead of falling back on local wandb.init() calls. The agent inspects launchable code and queue availability first. It then runs a minimal smoke test before touching any compute budget.
Trials execute one after another. The agent inspects metrics and hardware utilization from the completed job before picking the next configuration. Research progress stays in a local markdown file and syncs to a W&B Artifact via autoresearch_state.py when approved.
Python packages must be installed explicitly using uv. Without an active W&B Launch queue with verified capacity, the agent will refuse to schedule trials.
What it helps with
- verify code readiness and W&B Launch queue capacity
- smoke-test training workloads before full execution
- execute serial parameter trials based on observed metrics
- persist and resume experiment state using W&B Artifacts
Example prompts for the agent
- Inspect queue capacity and smoke-test our training job in W&B Launch
- Run serial trials testing learning rates on W&B Launch within a budget of 4 runs
- Save current autoresearch state to W&B Artifacts
How to install wandb-autoresearch
1Skills CLI (Vercel)
npx skills add wandb/skills --skill wandb-autoresearch -a claude-code2Manually: put the files in place
~/.claude/skills/wandb-autoresearch/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/wandb-autoresearch/
1Skills CLI (Vercel)
npx skills add wandb/skills --skill wandb-autoresearch -a codex2Manually: put the files in place
.agents/skills/wandb-autoresearch/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add wandb/skills --skill wandb-autoresearch -a cursor2Manually: put the files in place
.cursor/skills/wandb-autoresearch/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install wandb/skills skills/wandb-autoresearch2Manually: put the files in place
.github/skills/wandb-autoresearch/SKILL.md1Skills CLI (Vercel)
npx skills add wandb/skills --skill wandb-autoresearch -a gemini-cli2Platform CLI
gemini skills install https://github.com/wandb/skills --path skills/wandb-autoresearchManually: put the files in place
.devin/skills/wandb-autoresearch/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add wandb/skills --skill wandb-autoresearch -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about wandb-autoresearch
How to install wandb-autoresearch in Claude Code?
Run npx skills add wandb/skills --skill wandb-autoresearch -a claude-code in your terminal.
What does wandb-autoresearch do?
It enables coding agents to test training hypotheses on W&B Launch through verified queues, smoke tests, serial runs, and metric evaluation.
Does it run experiments in parallel?
No, trials run serially by default so the agent can inspect metrics between jobs. Parallel compute requires explicit user approval.
Source text
SKILL.md · Apache-2.0
--- name: wandb-autoresearch description: "Run bounded, evidence-driven training research through W&B Launch: assess project readiness, establish launchable code and queue capacity, smoke-test real jobs, execute serial trials, compare metrics, and persist resumable research state. Use when a coding agent is asked to autonomously test training hypotheses or tune a real W&B-tracked workload." --- <!-- SPDX-FileCopyrightText: 2026 CoreWeave, Inc. SPDX-License-Identifier: Apache-2.0 SPDX-PackageName: skills --> # W&B autoresearch Use W&B Launch to execute real training experiments, read their results, and choose the next bounded trial. Do not create a local `wandb.init()` run as a substitute for compute that was supposed to run through Launch. The Launch implementation lives in `skills/wandb-primary`: - `skills/wandb-primary/scripts/launch_helpers.py` - the `Launch` section in `skills/wandb-primary/SKILL.md` Install missing packages explicitly with `uv`; do not assume a fixed working directory or preinstalled environment. ## Flow Read only the reference for the current stage: | Stage | Entry condition | Read | |---|---|---| | Understand | Always first; inspect launchable code, queue capacity, data location, and prior state | `references/understand.md` | | Set up | Launchable code or usable queue capacity is missing | `references/setup.md` | | Search | Launchable code and queue capacity both exist | `references/search.md` | The readiness gate is `launchable code + queue with usable capacity`. A dataset may be a versioned Artifact or may be supplied by the code/image; its storage form alone does not determine readiness. ## Experiment records Use each W&B object for one job: - Runs hold trial config, metrics, system metrics, status, and code/job lineage. - Job or code Artifacts make code relaunchable. - Dataset Artifacts provide versioned inputs when the workload needs them. - Launch queues and agents provide compute. - A small local markdown file holds the hypothesis, trial tally, decisions, queue, and next step. Keep local state at a user-visible path in the current project or a path the user supplied. Sync it to a W&B Artifact only when cross-session resume is useful and the user authorizes the write: ```bash S=skills/wandb-autoresearch/scripts/autoresearch_state.py uv run --with wandb python "$S" load ENTITY PROJECT --dest . uv run --with wandb python "$S" save ENTITY PROJECT --path AUTORESEARCH_STATE.md ``` ## Compute safety - Confirm the queue, metric direction, search axis, and trial/compute budget before the first non-smoke launch. Do not exceed the agreed budget. - Smoke-test the real job at a deliberately small setting before full trials. - Change only config fields the program actually reads. Use a code edit and a new job Artifact for architecture or pipeline changes. - Run trials serially unless the user explicitly authorizes parallel compute. Inspect each completed result before selecting the next trial. - Report failures, hardware utilization, and uncertainty. Re-run a likely winner with another seed when the apparent gap could be noise. - Never fabricate a run, metric, queue result, or completion state.
© wandb · Apache-2.0 · wandb/skills