v1.108.263 - refresh --json was dead on arrival

jcodemunch-mcp refresh --json raised UnboundLocalError and printed nothing. It has been broken since the flag shipped in v1.108.259.

print(_json.dumps(_out, indent=2))     # _json is not bound at this point

Other handlers in main() do import json as _json further down the function. That makes _json a local for the whole of main(), so a branch dispatched above those imports fails with UnboundLocalError rather than NameError — which is also why copying the line from a neighbouring handler looked right.

Fixed by using the module-level json import.

The part worth writing down

Ruff reported this as F821 on every one of the four releases that carried it. The lint job was red on v1.108.259, .260, .261 and .262, all of which were committed, tagged, uploaded to PyPI and announced anyway.

Two independent gaps let that happen:

• Nothing in the test suite touched the CLI. tests/test_refresh_campaign.py exercised run() and status() directly, one layer below the defect. Testing the function you wrote instead of the command a user types is how an entire flag ships dead with 32 tests passing over it.
• The local suite is not the build. pytest does not run ruff, the 8-job test matrix passed the whole time, and the release flow never read the check after pushing. Every signal that was being watched was green.

Both are now closed rather than noted. The release checklist gains an explicit uv run ruff check src/ step and an explicit "read the CI run for the pushed SHA" step, and the tests now drive the real CLI in a subprocess: five cases covering --status --json, --json, the text control, a non-zero exit on refusal, and the UnboundLocalError signature by name. Four of the five fail before the fix.

Suite: 7337 passed, 7 skipped, 0 failed. Ruff: clean. CI: all 9 jobs green, verified on the pushed SHA before this release was created.