open-agent-ai-security/observra ? reverse-engineered prompt

Build me an open source Python library that helps me see what my AI agents are actually doing while they run.

I want it to work with a few popular agent frameworks, and I should be able to turn it on with just a small setup change, not add logging all over my code. It should capture the important stuff like model calls, tool use, token counts, cost, errors, session and trace context, and save everything in a clean structured format that I can send to a local file, a webhook, or an observability backend.

Please include built in support for redacting sensitive data, basic prompt injection detection, and simple health stats so I can tell if the telemetry pipeline is working. Make the output easy to plug into analytics or security tools, and include docs plus a few working examples for setup and usage. If you need current docs for any framework or backend detail, look them up online first.