메뉴
Triage a failing or misbehaving compiler test. Use this at the start of any debug session when a test is failing, a signal has a wrong value, a simulation mismatch appears, or a pass starts regressing and you need to choose the right artifact and workflow first.
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/miguel9554/mateIR --skill debug명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
Run the repo's canonical build and test workflows. Use this when validating compiler changes, running one test, running regression, or understanding expected-failure tests. Prefer the existing Makefile and tests/regression.py flows.
Use this skill whenever the user wants to plan a large refactor, migration, or multi-step code change that will be executed across multiple agent sessions with fresh context. Triggers include "plan this in phases", "create a phase plan", "break this into agent phases", "I want to do this step by step across sessions", or any time the user describes a large change and asks Claude to structure it for incremental execution. Produces a PLAN.md and a get_prompt.py script in a plans/ folder.
Inspect DFG .json or .dot debug artifacts produced by the compiler. Use this when debugging signal drivers, dependencies, fanout, op behavior, pass-to-pass changes, dead code, or node identity questions. Prefer tools/dfg_inspect.py and correlate with runtime tracing when structure alone is not enough.
SOC 직업 분류 기준
SKILL.md 표시 중
| name | debug |
| description | Triage a failing or misbehaving compiler test. Use this at the start of any debug session when a test is failing, a signal has a wrong value, a simulation mismatch appears, or a pass starts regressing and you need to choose the right artifact and workflow first. |
| Symptom | Start here |
|---|---|
| custom-sim vs Verilator output mismatch | read-vcd - find the first diverging signal or scope, then use trace-dfg-runtime or read-dfg on that cone |
| Signal has wrong or unexpected value | read-dfg - node --details, then deps, uses, or neighborhood |
| Need runtime evidence for how an op evaluated | trace-dfg-runtime |
| Entire struct or group of signals is zero | read-dfg - const_driven 0, then group or uses |
| Flop missing, wrong domain tag, or wrong reset | read-hierarchy - flops <path>, then check 06_flop_resolve_flops.txt |
| Module structure wrong (missing port, wrong hierarchy) | read-hierarchy - tree, then module <path> |
| Clock/reset domain error, CDC failure, domain propagation issue | debug-domains |
| Pass regression (worked before, broken now) | read-dfg - compare adjacent pass JSONs with diff; narrow to one node or cone when possible |
Artifacts live under the work directory used to reproduce the failure:
tests/<name>/work/static/debug_output/<top>/tests/<name>/work/custom-sim/debug_output/<top>/tests/<name>/work/custom-sim/output/Key files:
NN_<pass_name>.json and .dothierarchy.json06_flop_resolve_flops.txt11_domains_propagate_flops.txt or 12_domains_propagate_flops.txtoutput/dfg_trace.jsonl when runtime DFG tracing is enabledoutput/ or verilator/When the symptom is a wrong value or wrong structure:
trace-dfg-runtime.read-dfg - static DFG artifact inspection with node, nodes, deps, uses, neighborhood, and structural difftrace-dfg-runtime - runtime DFG evaluation tracing with --trace-dfg-node, --trace-dfg-cone, and --trace-dfg-opread-hierarchy - hierarchy.json inspection (tree, flops, find, clocks)read-vcd - waveform inspection and unified cross-sim diff (vcd_inspect.py, vcd_diff.py)debug-domains - domain/CDC pass workflow and sidecar schema