Probe live agent trajectories and write findings reports in the Varro playground.
npx skills add https://github.com/josca42/varro --skill playground-explorer
Please help me install this Agent Skill. Skill: playground-explorer Source: https://github.com/josca42/varro/tree/main/.agents/skills/playground-explorer Command: npx skills add https://github.com/josca42/varro --skill playground-explorer
Live interactive trajectory exploration and improvement of agent behavior in the Varro playground, enabling researchers to probe observations and actions as they unfold and to propose concrete improvements to tools, prompts, docs, and UI-as-agent-tools.
Launch a CLI probe by running uv run python -m varro.playground.cli and begin a subject exploration.
High-intent search queries and answers about installing and using this skill.
To debug agent trajectories interactively in the Varro playground, launch a CLI probe by running uv run python -m varro.playground.cli to inspect live observations and trace decisions.
Interactive trajectory analysis is the process of probing live observations and trajectory artifacts to root-cause agent failures and propose concrete improvements to prompts and tools.
You diagnose agent decisions by inspecting trajectory artifacts including turns, tool_calls, and images via CLI-based probes to trace exactly where and why the agent failed.
You generate a findings report for prompt improvement by proposing changes during exploration, which the system writes to data/trajectory/{user_id}/{chat_id}/findings.md.
Yes, you can propose concrete improvements to UI components, as well as prompts and docs, while probing agent behavior in the Varro playground.
No external dependencies are required to run CLI probes for trajectory exploration, as the Skill operates independently to generate trajectory data under data/trajectory.
SstoboInteract with Convex agents for testing and debugging in a graphical playground.
DieStokExecute end-to-end browser tests on pages modified by a PR or branch.
mmbianco78Trace software errors backward to their root cause before deployment.
yo-stevenDiagnose failing NemoClaw sandboxes via health, logs, and live network activity.
