[rahul05ranjan]

Fixes #854.

What was wrong

The Claude Code install and uninstall paths (HookServer.swift) removed a whole hook group from ~/.claude/settings.json when any command inside it contained the substrings coucou or NotchBuddy:

• buildHooksData() (~L1481) — the per-event cleanup before reinstalling;
• uninstallClaudeHooks() (~L1506);
• the App Store variants buildHooksData(claudeURL:) (~L1719) and uninstallClaudeHooksAppStore(claudeURL:) (~L1682).

So a user hook like ~/dev/coucou/scripts/lint.sh (likely for contributors to this repo), or any user hook sharing a matcher group with one of ours, was deleted on every install or uninstall. The Gemini path never had this problem: removeNbHookEntries (~L2011) removes only the entries whose command contains nb-hook, and drops a group only when every entry in it was ours.

The fix

All four Claude sites now call the same removeNbHookEntries the Gemini and Codex paths already use. Both builds' own commands always contain nb-hook (<supportDir>/nb-hook non-App Store, <~/.claude>/coucou/nb-hook App Store, incl. the --ask and status-line variants), and nb-hook is already the marker the rest of the file uses for "this one is ours" (statusLineInstalled, codexHooksInstalled, the Codex and Gemini cleanup). The doc comment on removeNbHookEntries now says it serves both formats.

Also added NotchBuddy/Tests/ClaudeHookRemovalTests.swift, a standalone script in the HermesConfigMergerTests.swift style:

swift NotchBuddy/Tests/ClaudeHookRemovalTests.swift

It keeps a copy of removeNbHookEntries plus the per-event install cleanup in sync, and covers: a user ~/dev/coucou/... group surviving install, no duplicate nb-hook entries on reinstall, echo coucou surviving uninstall while the nb-hook entry goes, coucou-only groups dropped entirely, a /opt/NotchBuddy/... path that is not our hook surviving uninstall, and the legacy flat format.

Verification

No Swift toolchain on this machine (arm64 Linux CI box), so:

• Scenario harness, Python 3 mirror of the exact before/after logic (group-level removeAll on the two substrings vs per-entry nb-hook removal), run on the settings shapes from Installing/uninstalling hooks removes user hooks whose command contains "coucou" #854:

scenario	before	after
install with a ~/dev/coucou/scripts/lint.sh group	user group deleted	user group kept, fresh nb-hook group appended
uninstall of a matcher holding echo coucou + our nb-hook	whole matcher deleted	echo coucou kept, nb-hook entry removed
uninstall of a coucou-only event key	key removed	key removed (unchanged)
reinstall twice	one plain nb-hook group	one plain nb-hook group, one AskUserQuestion group, no duplicates
user hook at /opt/NotchBuddy/something-else	deleted	kept
user hook named ~/dev/coucou/... on macOS App Store paths (lowercase coucou)	deleted	kept

• Read-back of the live flows on HEAD before the change: grep -n 'NotchBuddy.*removeAll\|coucou.*removeAll' NotchBuddy/Sources/App/HookServer.swift found the four group-level sites and nothing else; the status-line path already used cmd.contains("nb-hook") ("not ours — leave unchanged"), which is the behavior this PR extends to the group paths.
• I could not run swift NotchBuddy/Tests/ClaudeHookRemovalTests.swift locally (no toolchain); the script follows the same copy-and-assert pattern as the existing HermesConfigMergerTests.swift, so it runs the same way.

Two deliberate non-changes: the detection reads (hooksNeedUpdate, claudeHooksInstalled) still match the broad substrings — they only answer "are our hooks installed?" and removing nothing; and a user hook whose command itself contains nb-hook would still be treated as ours, exactly as in the Gemini and Codex paths today.