fix(cli): the update check honours SURREAL_MEMORY_NO_UPDATE_CHECK, and the suite sets it

[RobertSigmundsson]

Summary

• run_update_check_background() gains an opt-out: it short-circuits when SURREAL_MEMORY_NO_UPDATE_CHECK is truthy.
• tests/conftest.py sets it for the session, so a full test run makes no calls to PyPI.
• Truthiness follows unified_config._env_truthy, the convention this repo already uses everywhere else.
• Adds nineteen tests over the accepted spellings, the rejected ones, and the unset case.

Why

The update check had no opt-out at all. Most CLI invocations start a daemon thread that queries PyPI — the callback skips a short list of commands and --json, but not the several hundred invocations a full test run makes. On a machine with egress that is real outbound traffic from a unit test suite; on a machine without one it is the connection timeout instead.

It also feeds a flake, though not in the way it first appears. The banner goes to stderr, deliberately, with a comment in the code explaining that it must not break a piped smem recall … | jq. But CliRunner merges stderr into result.output, and the CLI tests parse that as JSON — so the banner lands in a different test's parse on each run.

Measured with strace -f -e trace=connect, counting connections to anything other than loopback, with a clean HOME so the 24-hour cache cannot mask the result:

tree	SURREAL_MEMORY_NO_UPDATE_CHECK	non-loopback connections
main	unset	10
main	1	10 — the variable does not exist yet
this branch	unset	10 — the check still works
this branch	1	0
this branch	on, TRUE	0
this branch	off, 0	10

The third row is the one that matters as a control: the opt-out suppresses the check, it does not break it.

Why this truthiness and not the obvious one

The obvious implementation — "any non-empty value disables it" — would give SURREAL_MEMORY_NO_UPDATE_CHECK=off the effect of switching the check off, which is the opposite of what anyone typing that would expect, and it would be a fourth convention in a repo that already has one.

So the gate calls unified_config._env_truthy, the same helper behind SURREAL_MEMORY_EMBEDDING_ENABLED, SURREAL_MEMORY_SYNC_ENABLED, SURREAL_MEMORY_SYNC_AUTO and the three SURREAL_MEMORY_REASONING_* switches: 1/true/yes/on, case-insensitive, and everything else — off, no, 0, false included — leaves the check running. The last two rows of the table above are that decision, measured.

Changes

• cli/update_check.py: the short-circuit and a docstring recording the convention and why the test suite sets the variable.

• tests/conftest.py: _isolated_home_dir sets SURREAL_MEMORY_NO_UPDATE_CHECK=1 alongside the $HOME redirect it already installs. Issue Test suite overwrites the user's real claude_desktop_config.json #110 and PR fix(tests): stop unit tests from writing into the real Claude Desktop config #121 closed the write side of that fixture; this closes the network side. A test that needs the real code path can monkeypatch.delenv it, and tests/unit/test_update_check_env_gate.py does exactly that.

• tests/unit/test_update_check_env_gate.py (new).

• docs/reference/config.md: regenerated with scripts/gen_config_docs.py, since that file is generated from these defaults and the Docs Freshness job checks it.

Test plan

• pytest tests/unit/test_update_check_env_gate.py — 19 passed.
• The strace table above: measured on both trees with a fresh HOME each time. The first attempt at this probe reported zero connections everywhere, because the check caches its answer for 24 hours in ~/.surrealmemory/.update_check and the real HOME had a warm cache — worth mentioning in case anyone reproduces it and wonders why nothing happens.
• pytest tests/ -m "not stress" -n 4 against a live SurrealDB v3.2.0 — 7302 passed, 48 skipped, 1 xfailed, which is main's 7283 plus exactly the nineteen tests added here. Two tests in tests/unit/test_dashboard_brains_scope.py fail on this branch and on main alike: they want a live database and collide with one another under -n. Both pass when that file is run on its own.
• ruff check src/ tests/ clean; ruff format --check src/ tests/ reports 740 files already formatted.
• mypy src/ --ignore-missing-imports — success, no issues found in 354 source files.
• Coverage under the CI gate: 72.30%, against 72.36% on main.
• CHANGELOG.md untouched — left to the release entry, as with fix(storage): bind datetimes in time comparisons so they select by value #191–fix(memory): refresh content-derived fields on compress, restore, and refine #193.

On the flake

The JSON-parsing flake in the CLI tests comes from the same banner reaching result.output through CliRunner, and setting the variable for the suite removes the banner. I am reporting that as a consequence rather than as this PR's proof: a green run is not by itself evidence that a flake is gone.

Verified by

@RobertSigmundsson

[acidkill]

Reviewed the full diff. Env gate follows the repo's canonical _env_truthy convention, conftest sets it session-wide, and falsy values (off/no/false) deliberately do not disable — an env var that looks like a switch never silently disables the check.