test: cover renderer separators, ANSI edge cases and settings errors - #660

What

Adds 65 tests for three core utilities. Test files only; no production code changes.

• src/utils/renderer.ts
  • Powerline: separator (arrow) colors for every background combination (different, same, only the left widget, only the right widget, none), in normal and inverted mode; start, flex-boundary and end caps on widgets without a background; bold reset before a flex gap and after an end cap; bold kept through the separator after a bold, dimmed widget; widgets that keep their own colors (preserveColors); auto-align padding for merged groups, and merges across or into widgets that rendered empty.
  • Standard mode: separator widgets and the automatic defaultSeparator taking the previous widget's colors with inheritSeparatorColors (including its default color), the background override on separators, , and - separator spacing, padding colored with the widget background on each padding side, the preview width in full-minus-40, and widget types this version doesn't know.
• src/utils/ansi.ts (new ansi.test.ts): 8-bit C1 CSI/OSC sequences, non-hyperlink OSC (window titles), unterminated CSI/OSC, two-character ESC x escapes, flags, ZWJ emoji sequences, stray combining marks, and truncation that never splits a flag or ZWJ sequence and closes hyperlinks correctly.
• src/utils/claude-settings.ts: loadClaudeSettingsSync, getRefreshInterval / setRefreshInterval against unreadable settings, a failed .orig backup during install, invalid or malformed ccstatusline configs during install, and getRemoteControlStatus.

Why

• Powerline separator coloring had tests for two of its ten branches (normal mode, different backgrounds or none). The rest pick the arrow color whenever neighbors share a background, only one has a background, or the separator is inverted. A mistake there shows up as an invisible arrow or a hole in the color band.
• The bold and dim resets at flex gaps and end caps keep bold from leaking into the next segment, or past the end of the line into Claude Code's UI.
• inheritSeparatorColors (for separator widgets and the automatic defaultSeparator) and the background override on separators and padding had no tests. If they break, a configured color band breaks at every separator and padding space.
• The ANSI parser measures custom-command output, which can contain anything a script prints: C1 controls, cursor save/restore, hyperlinks, emoji. A width error there misplaces flex separators and truncates at the wrong column.
• getRemoteControlStatus, behind the Remote Control widget, had no direct tests (the widget's tests mock it). How it handles a missing sessions directory, non-manifest files, malformed manifests and other sessions was unverified.
• installStatusLine must still install when the .orig backup can't be written, and must not sync hooks from a ccstatusline config that fails validation or isn't valid JSON.

How

• Three new focused files: ansi.test.ts, renderer-powerline-separators.test.ts and renderer-separator-colors.test.ts. Everything else extends the existing test file for that topic (renderer-dim, renderer-flex-width, renderer-padding-side, renderer-separator-collapse, claude-settings).
• Expected strings are hand-written literals. Colors are hex: values, which produce raw truecolor codes whatever chalk detects about the terminal. The one place that needs a widget's named default color pins chalk.level and restores it afterwards, as renderer-config-warning.test.ts already does.
• Renderer tests pass terminalWidth explicitly (0 for "width unknown"), so nothing depends on the terminal running the tests.
• File-system tests use real temp directories under CLAUDE_CONFIG_DIR, like the rest of claude-settings.test.ts. A failed backup write and an unreadable session manifest both come from a directory sitting where the file should be. Nothing spies on fs, child_process or os.
• The "no manifest matches" test puts every kind of skipped entry in one directory without a match, so each skip branch runs whatever order readdir returns.

Left uncovered on purpose:

• claude-settings.ts: isNpmAvailable, isNpxAvailable, isBunAvailable, isBunxAvailable and getPackageCommandAvailability, which run which / where in a child process; and the Windows (cmd.exe) branch of the --config path quoting, which needs process.platform mocked to win32.
• renderer.ts: the fallback for an unknown flexMode (settings are validated against the three modes on load), the catch around widget rendering (nothing inside throws for a registered widget), and the default-separator branch for an element without a widget (every element has one).
• ansi.ts: the fallback for runtimes without Unicode property escapes, guards for indices that can't be out of range, and truncation's "nothing was cut" return, which needs an escape sequence inside a single emoji cluster.

Demo

No demo: test-only, nothing visible changes.

Testing

• Line coverage, full suite with bun test --coverage:
  • src/utils/renderer.ts: 90.41% → 99.35%
  • src/utils/ansi.ts: 84.54% → 98.31%
  • src/utils/claude-settings.ts: 81.90% → 95.81%
• bun test: 2423 pass, 0 fail, 151 files (main: 2358 pass, 148 files), so 65 new tests.
• bun run lint: clean.
• Each of the 8 touched test files run 15 times in a row under Bun (120 runs): all passed.
• Node 26.10.0, node node_modules/vitest/vitest.mjs run on the 8 touched files: 230 tests, 219 pass, and all 65 new tests pass. The 11 failures are the existing execSync spies in claude-settings.test.ts, which fail the same way on main.
• Mutation spot checks, each reverted afterwards: disabling flag pairing, ZWJ joining or C1 string-terminator parsing in ansi.ts; using the wrong text color for the inverted same-background arrow, dropping the bold reset before a flex gap, or dropping the 22;1 bold restore in renderer.ts; ignoring the previous widget's background on inherited separators; dropping the placeholder for unknown widget types; and removing the empty-session-id guard, the .json filter or the schema check before hook sync in claude-settings.ts. Each one made at least one new test fail.