[KoenVdH]

Problem

Running the test suite (npx vitest run) overwrote the developer's real ~/.config/openhamclock/rig-bridge-config.json with defaults: radio none, empty API token, serial port gone.

The cause is that plugins/brightness.js and plugins/aprs-tnc.js saved settings with require('../core/config').saveConfig(). That always writes the core config object to the real config path, whatever config the plugin was created with. The brightness tests in hardware-plugins.test.js create the plugin with their own config and change a setting, so each run replaced the user's file with the core defaults.

Fix

• Plugins: brightness and aprs-tnc now use the saveConfig the core already passes to every plugin (services.saveConfig, rig-bridge.js), like gpio does. Without it they skip saving instead of falling back to the real file. Production behaviour is unchanged.
• core/config.js: a new RIG_BRIDGE_CONFIG_DIR environment variable overrides the config location outright, with no legacy lookup or migration.
• src/test/setup.js: sets RIG_BRIDGE_CONFIG_DIR to a temp dir for every test run, so even a future test that saves config can't reach the real file.

Tests

• brightness: saves through the injected saveConfig only when a setting changes, and works without it.
• aprs-tnc: setBeaconEnabled writes through to the config and saves through the injected saveConfig, and doesn't throw without it.
• core/config.test.js: under test, CONFIG_DIR/CONFIG_PATH point at the temp dir, not ~/.config/openhamclock.

Verified

Full suite run with a scratch HOME: 310 files and 3,776 tests passed. Nothing was written to the scratch HOME, where the brightness test used to write. The real config file's checksum and modification time were unchanged. eslint and prettier are clean.

🤖 Generated with Claude Code

[accius]

Thanks, Koen. Good diagnosis: require('../core/config').saveConfig() always writes the core config object, so any plugin created with its own config overwrote the real file. Using the saveConfig the core already passes to plugins (rig-bridge.js line 86), and skipping the save without it, is the right fix. The RIG_BRIDGE_CONFIG_DIR setup guard is a nice extra layer.

I verified it on a machine that had been bitten. Test-merged onto current Staging, I ran the full suite including hardware-plugins.test.js (310 files, 3,775 tests pass). The real ~/.config/openhamclock/rig-bridge-config.json had the same modification time and SHA-1 before and after. Approving.