Tool loop guard stopped repeated failing calls to "read" after 3 attempts (limit 2). Adjust tool arguments and retry.

[ZqinKing]

The model provided by cursor-acp often breaks down due to the problem described in the English title.

[brunoamui]

This may be caused by the same root cause identified in #48 — the local bash/shell tool handler uses child_process.exec() which throws on non-zero exit codes and has a timeout unit mismatch (model sends seconds, exec expects milliseconds). This causes most tool calls to fail with SIGTERM or non-zero exit errors, which then triggers the loop guard.

[Nomadcxx]

Root cause fixed in f5fc7b5 via #48. The bash tool was killing commands after ~5ms and throwing on non-zero exits, causing the loop guard to correctly detect repeated genuine failures. Fix: switched to spawn(), corrected timeout unit conversion (seconds→ms), added user shell.

[Nomadcxx]

Should be fixed in 2.3.17 — the bash tool was killing commands after ~5ms due to a timeout unit bug, which is what was causing the loop guard to trip. Update with bun add @rama_nigg/open-cursor@latest and let me know if you're still seeing it.