---

name: public-api-and-automation description: "Query, export, and automate W&B resources with wandb.Api, including runs, projects, sweeps, reports, files, history exports, and automation events/actions/scopes." disable-model-invocation: true metadata: disco-role: operating license: MIT

W&B Public API and Automation

Use this sub-skill when a task asks an agent to read existing W&B data, export run history or metadata, inspect files/reports/sweeps, or create/update W&B Automations. This is for post-hoc querying and server-side automation, not live experiment logging, artifact lifecycle design, or running sweep agents.

Route by task

• For wandb.Api construction, path shapes, run/project/sweep/report access, filters, and object properties, read API reference.
• For large or selective run-history exports, sampled vs unsampled history, pagination, file downloads, and the bundled export helper, read export and pagination.
• For automation scopes, events, filters, integrations, actions, create/update/delete flows, and compatibility checks, read automations.
• For auth, host, path, pagination, missing-field, GraphQL/network, and automation mismatch failures, read troubleshooting.

Core patterns

import wandb

api = wandb.Api(timeout=60)
run = api.run("entity/project/run_id")
print(run.config)
print(dict(run.summary))
for row in run.scan_history(keys=["loss", "accuracy"], page_size=1000):
    ...

import wandb
from wandb.automations import OnRunMetric, RunEvent, SendWebhook

api = wandb.Api()
project = api.project("project", entity="entity")
webhook = next(api.webhook_integrations(entity="entity"))
event = OnRunMetric(scope=project, filter=RunEvent.metric("loss").mean(5).gt(1.0))
action = SendWebhook.from_integration(webhook, payload={"metric": "loss"})
automation = api.create_automation(event >> action, name="loss-alert")

Bundled script

Use scripts/export_run_history.py for safe CSV or JSONL exports:

python scripts/export_run_history.py --entity ENTITY --project PROJECT --run RUN_ID --out history.csv --format csv --keys loss accuracy --max-rows 10000

Start with --dry-run or --max-rows before exporting large histories.