Executes live UI actions on the user's open LangWatch page via CLI commands.
npx skills add https://github.com/langwatch/langwatch --skill drive-the-ui
Please help me install this Agent Skill. Skill: drive-the-ui Source: https://github.com/langwatch/langwatch/tree/main/skills/_compiled/native/drive-the-ui Command: npx skills add https://github.com/langwatch/langwatch --skill drive-the-ui
When a user has a LangWatch page open (such as the evaluations workbench), changes made behind their back leave their screen stale and confusing. This Skill lets the agent operate the page directly in the user's browser so edits happen visibly in front of them, with a backend fallback when no browser answers.
langwatch ui actions lists the action kinds the current page accepts, with JSON schemas and required permissions, so payloads are never guessed.langwatch ui call <kind> executes one typed action and reports executedVia as browser (user saw it happen) or backend (applied to saved state).langwatch workbench get-state <experiment-slug> reads live workbench state including unsaved prompt drafts and in-memory results.Ask the agent to list the available UI actions on the current page and then update the target prompt on your open evaluations workbench so you can watch the change happen live.
High-intent search queries and answers about installing and using this skill.
Run `langwatch ui actions` first to list the action kinds the current page accepts with their payload schemas, then call one with `langwatch ui call <kind> --payload-file <path>`. The action executes in the user's browser and the result returns in the same command.
Write the JSON payload to a file and use `--payload-file <path>`, or pipe it via stdin with `-`. Inline `--payload` quoting breaks on apostrophes in prose, causing the command to refuse the payload.
`browser` means the user's open page applied the action and they watched it happen. `backend` means no page answered, so the platform applied it to the saved state and the user's page is a step behind until it catches up.
The payload did not match the action's JSON schema. Read `meta.issues` in the error response, fix the payload accordingly, and retry once. Run `langwatch ui actions` first to confirm the expected shape.
No. UI actions run mid-turn only; outside a turn the call is refused with `langy_ui_turn_inactive`. They also require the LangWatch CLI inside a Langy worker session.
Do not retry blindly, because the page claimed the action and may have half-applied it. Re-read the state with `langwatch workbench get-state <experiment-slug>` first, then decide whether to retry.
qvidal01Manage and automate Cloudflare Workers, Pages, storage, AI, networking, and security features.
selcukyucelOptimize skill and agent prompts through an automated evaluation loop.
JingyiWngOrchestrate end-to-end GenAI agent evaluation with MLflow datasets, scorers, and tracing.
DevFlex-AIEnforce Agent Sessions layout guidelines to keep code aligned with sessions/LAYOUT.md.