[dustinbyrne]

Motivation and context

Games need to start session replay after a gameplay milestone rather than immediately after SDK setup.

• Add UPostHogSubsystem::StartSessionReplay() for C++ and Blueprints.
• Add Replay.bAutoStart (default true). Set it to false with Replay.bEnabled = true to defer recording until an explicit call after setup.
• Repeated starts reuse the recorder. Existing consent, remote configuration, foreground/session gates and shutdown ownership remain unchanged.

Targets feat/rust-v1-embedding, which contains the experimental replay implementation. This is an additive API with unchanged defaults; no migration or native ABI change is required. Remote-config endpoint support remains separate work.

How did you test it?

On macOS arm64 with Unreal Engine 5.8:

• Built the native bridge with scripts/build-native.sh and the base branch's local Rust embedding dependency, then built the Unreal consumer with scripts/build-example.sh.
• Ran Tests/native_integration.py against the maintained SDK mock server: 12 native integration tests passed.
• Ran its Unreal editor integration test: all 9 PostHog.* automation tests passed under -nullrhi.
• Ran PostHog.ReplayStartup again with rendering enabled: verified the reflected Blueprint entry point, no startup in manual mode, idempotent successful start, automatic startup, and shutdown. Two real recorder instances started and stopped; remote capture permission remained closed.
• New tests cover default/options conversion and rejection before setup, while initializing, without local replay opt-in, and during shutdown.
• Independent read-only review found no issues. git diff --check passed.

Checklist

• I reviewed the submitted code.
• I added or updated tests where applicable.
• I updated the documentation where needed.
• I described any breaking changes and their migration path.

🤖 Agent context

Autonomy: Human-driven (agent-assisted)

Implemented with Pi using file-editing and shell tools, plus an independent read-only reviewer. The existing recorder remains owned by the game-instance subsystem; automatic and manual startup share the same implementation and capture gates. Requires human review.

[marandaneto]

The code looks good; no qualifying findings. Verdict: correct.