Repository navigation
rtk init claims any Claude or Cursor hook command that mentions rtk-rewrite.sh #4427
Description
Claude and Cursor recognise the legacy hook with cmd.contains("rtk-rewrite.sh"): hook_already_present and remove_legacy_hook_entries_from_json in src/hooks/init/claude.rs, is_cursor_hook_entry and remove_legacy_cursor_hook_entries_from_json in src/hooks/init/cursor.rs. A user hook that only mentions the name is treated as RTK's.
Repro on develop 6f5503d, with ~/.claude/settings.json holding one PreToolUse Bash group:
[{"type":"command","command":"echo rtk-rewrite.sh"},
{"type":"command","command":"/custom/rtk-rewrite.sh.backup"},
{"type":"command","command":"python3 /custom/audit.py --label rtk-rewrite.sh"}]rtk init -g --auto-patchprintssettings.json: hook already presentand registers nothing; the file is unchanged, so RTK is not installed.rtk init -g --uninstallremoves all three (settings.json: removed RTK hook entry).- The same uninstall, with
~/.cursor/hooks.jsonholdingecho rtk-rewrite.sh,/custom/rtk-rewrite.sh.backupandmy-guardunderpreToolUse, leaves onlymy-guard.
Expected: only the script run on its own (one word whose file name is rtk-rewrite.sh) is RTK's. The Codex installer is getting the same rule in #4309.
Related: #4227 (the same class for Gemini), #3036 (uninstall removes a whole matcher group), #3693 (legacy script deleted before the replacement is registered).
Activity
rtk-wshm-sync-bot commentedon Oct 3, 2026
🔍 Automated Triage
bugmediumSummary
Claude/Cursor hook detection uses substring match on 'rtk-rewrite.sh', wrongly claiming unrelated user hooks that merely mention the name during init/uninstall.
📁 Relevant files
src/hooks/init/claude.rssrc/hooks/init/cursor.rsSuggested Actions
hook_already_present/remove_legacy_hook_entries_from_jsonin src/hooks/init/claude.rs to match the command as a…is_cursor_hook_entry/remove_legacy_cursor_hook_entries_from_jsonin src/hooks/init/cursor.rsrtk init --geminitreats any BeforeTool group whose first hook contains "rtk" as RTK's: uninstall deletes user hooks, install skips #4227 (Gemini) since they share the same matching bugTriaged automatically by wshm · This is an automated analysis, not a human review.
TechWizard9999 commentedon Oct 5, 2026
Hi @KuSh , Could you please assign it to me
TechWizard9999 commentedon Oct 5, 2026
Hi @KuSh , About this implementation
We'd decouple the OpenCode plugin installation from Claude side failures: always attempt the OpenCode install, then report any errors, with regression tests for the malformed CLAUDE.md and read-only ~/.claude cases.Is this design okay ? or anything ekse
KuSh commentedon Oct 5, 2026
Hi @TechWizard9999, that's not the subject of the issue. Most of what you're talking about is already handled in #4269. If something is missing there, either review/comment on it or push a separate PR for what's missing.