clay-playbooks

Use the clay CLI: create the workflow with `clay workflows create`, add nodes with `nodes create`, pin inputs on each node's inputSchema with sourceNodeId and sourcePath, then publish and run through a routine. Tables cannot be created from the CLI and must be built in the browser.

The AI column's reasoning effort is at default, so the model spends its token budget on reasoning and returns empty content. Set reasoningLevel to the lowest accepted value in the UI dropdown; do not fix it by raising reasoningBudget, which costs roughly 35x more per row.

No. The CLI's tables command group is read-only (list, get, columns, rows, query). Tables and columns must be built in the browser, while workflows can be created, published, and run entirely from the terminal.

Pick one of the 19 signal playbooks, each producing a single output field like funding_line or hiring_line. Build it as a table column or CLI workflow, gate paid enrichment behind a valid email check, smoke test 5 rows, then push the field to your sequencer as a custom variable.

The node's inputSchema was dropped, either because inputSchema and outputSchema were sent in one update call or because deleting an upstream node nulled it. Send the schemas in separate calls and re-apply inputSchema on downstream nodes after any deletion.

The code node sandbox has no datetime module, no urllib, no outbound HTTP, and no pip; use time.strftime for dates. The run_code preview is not faithful since its sandbox includes datetime, so only a real workflow run proves a code node works.