Choose your framework
Each guide includes installation, credentials, a complete integration, and the framework-specific conversation and replay behavior.The integration pattern
- Enable tracing with your existing integration. Use a callback, adapter, or wrapper and pass the same PandaProbe client to Reflect.
- Attach Reflect to the native agent. Factory-based integrations use
reflect.agent(...). Some Python frameworks accept an existing agent or options object directly throughreflect.run(...). - Run through an explicit boundary.
reflect.run(...)captures the task’s traces, evaluates them, and invokes the supported reflection continuation. It returns the original task result. - Provide validation before publishing guidance. Supply fresh replay environments and an independent outcome verifier. Proposal generation alone does not publish a shared rule.
reflect_bash manually when using these integrations.
Shared behavior
- You keep your model, prompt, business tools, and native task result.
- Reflection exposes only Reflect filesystem administration through the supported native integration.
- Accepted rules are shared through the persistent workspace. Native conversation history and session notes are separate.
- Python configuration uses snake_case; TypeScript uses camelCase. Persisted records in both languages use snake_case.
- Keep complete stream consumption inside an explicit execution boundary when forwarding tokens to a UI.

