eytanlevit/human

Overview

This skill connects your agent to real humans for questions, verification, and expert advice. It provides an async API to submit prompts, track request status, and receive human responses without blocking your main flow. Use it when human judgment, approval, or testing is required.

How this skill works

Register once to receive an API key, then POST questions to the /v1/ask endpoint to create a request. The service returns a requestId and an initial status; responses are delivered asynchronously. Poll using /v1/wait for long-polling (recommended) or /v1/status for quick checks, or spawn a background waiter to notify users when a human reply arrives.

When to use it

• You’re stuck and need domain expertise or a sanity check.
• You need human approval before a destructive or high-risk action.
• You require real-world testing or UX feedback from a person.
• You want to verify model output against human judgment.
• You need clarification on ambiguous requirements or policy decisions.

Best practices

• Register once and securely store the returned apiKey for all requests.
• Use /v1/wait with a 30s timeout to reduce polling frequency and server load.
• Spawn a background waiter or separate task to avoid blocking user interactions.
• Inform users that replies are async and may take seconds to minutes.
• Retry polling for at least 5–10 minutes before giving up; handle 401/400/404 errors gracefully.

Example use cases

• Ask a human to confirm whether to deploy a breaking change or run a migration.
• Submit a UI flow or signup link for a human to test and report expected behavior.
• Request a technical sanity check on complex code or architectural decisions.
• Get a quick expert opinion when an automated system is uncertain.
• Request approval from a team member before executing a high-impact operation.