[eli-r-ph]

💡 Motivation and Context

main shipped 7.83.0 and 7.83.1 after #1174, so 8.0 would keep reading the /flags?v=2 response (#1031) and miss the MCP 2.x context fix (#1166). #1031 is the largest main change since the last sync, so it gets its own PR. Stacked on #1174.

Changes

• -x picks of the six main commits after chore: sync main into v1 (7.81.2–7.82.2) #1174, in order, release and reference commits included: feat(flags): read the /flags?v=4 response #1031, 7.83.0, fix(mcp): inject context into mcp 2.x tool objects and test without fastmcp #1166, 7.83.1 and their reference updates. None skipped.
• feat(flags): read the /flags?v=4 response #1031 conflicts, resolved toward v1:
  • async_client.py: v1's import lists, with feat(flags): read the /flags?v=4 response #1031's removals (_metadata_has_experiment, _parse_flag_payload, FlagMetadata) and additions applied.
  • request.py: v1's post() call, now to /flags/?v=4.
  • sdk-compliance.yml: v1's workflow kept. It runs harness 1.26.0 directly, newer than main's 1.23.0 pin.
  • Compliance adapter: flags_response_v4 added to v1's capability list, and asserted in v1's capability test.
• Every other pick changes exactly the lines upstream changed.
• One commit fits feat(flags): read the /flags?v=4 response #1031's new parity tests to v1: they stubbed _enqueue to read events, but v1 adds flag properties inside _enqueue, so the stub saw none and the 7.82.1 comparison failed. The tests now read the built event in before_send, so they compare v1's real output with 7.82.1 and pass. This also stops the v2/v4 comparison from passing with empty flag properties.
• Capture is unchanged: feat(flags): read the /flags?v=4 response #1031 changes only the /flags request and how its response is read. FlagMetadata.id and .version become optional, as on main.
• No migration guide change: no synced feature behaves differently on 8.0.

💚 How did you test it?

• Full pytest, ruff, mypy, check_public_api.py and uv lock --check locally: 6322 passed. The three metrics-autocapture fork tests fail on macOS, as on unmodified main (chore: sync main into v1 (7.75.7–7.81.1) #1169).
• feat(flags): read the /flags?v=4 response #1031's 169 flags tests pass, including the 7.82.1 parity scenarios.
• Each pick's +/- lines match its upstream commit, except the feat(flags): read the /flags?v=4 response #1031 resolutions above.

📝 Checklist

• All commits are signed (see Commit signing)
• I reviewed the submitted code.
• I added tests to verify the changes.
• I updated the docs if needed.
• No breaking change or entry added to the changelog.

If releasing new changes

• Ran sampo add to generate a changeset file

🤖 Agent context

Autonomy: Human-driven (agent-assisted)

• Written with Claude Code (claude-opus-5-5), following the release runbook's main-sync step and chore: sync main into v1 (7.75.7–7.81.1) #1169's method.

🤖 Generated with Claude Code

[turnipdabeets]

Picks match upstream. In #1031 the only differences are v1's sdk-compliance workflow (harness 1.26.0, which is newer than main's 1.23.0) and flags_response_v4 added to v1's capability list. The parity-test fix is correct: v1 reads self.before_send when the event is sent and adds flag properties inside _enqueue, so reading in before_send compares the real events. Nothing on main up to 7.83.1 was skipped.