What problem does it solve? Ascend C operators running on Ascend NPU hardware can hang without response, crash with Segmentation Fault or Abort, or fail intermittently due to memory corruption, and these failures are hard to localize without a systematic debugging workflow. ## Core Features & Use Cases - Symptom-based routing: A decision tree and symptom-cause table route hangs, timeouts, crashes, deadlocks, and aic errors to the right debugging workflow. - Coredump and plog analysis: Guides GDB backtrace analysis of core files and includes a parse_plog.py script that extracts timeout, crash, and hardware exception signals from Ascend plog logs. - Proactive memory error detection: Automates mssanitizer memcheck via run_memcheck_pre.sh and a JSON config template to detect illegal read/write, multi-core overwrite, misaligned access, illegal free, memory leaks, and unused memory. - Use Case: Your operator occasionally crashes with an unclear coredump stack and aic error in the logs; the skill walks you through compiling with sanitizer options, running memcheck, and interpreting the ERROR/WARNING report to locate the out-of-bounds access. ## Quick Start Ask the assistant to load the ascendc-crash-debug skill and diagnose why your Ascend C operator hangs or crashes on the NPU.
