菜单
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.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/miguel9554/mateIR --skill read-dfg命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
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.
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.
基于 SOC 职业分类
正在显示 SKILL.md
| name | read-dfg |
| description | 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. |
Use tools/dfg_inspect.py for JSON and dot -Tsvg for DOT.
Common locations:
tests/<name>/work/static/debug_output/<top>/tests/<name>/work/custom-sim/debug_output/<top>/Key idea:
trace-dfg-runtime when you need to know what happened at simulation time.Useful commands:
python3 tools/dfg_inspect.py <file> node <id-or-name> [--details]python3 tools/dfg_inspect.py <file> nodes [--op OP] [--name PATTERN] [--at FILE:LINE]python3 tools/dfg_inspect.py <file> deps <id-or-name> [--op OP] [--depth N] [--details]python3 tools/dfg_inspect.py <file> uses <id-or-name> [--op OP] [--depth N] [--details]python3 tools/dfg_inspect.py <file> neighborhood <id-or-name> --fanin N --fanout Npython3 tools/dfg_inspect.py <file> diff <other.json> [--node ID-OR-NAME] [--cone] [--details]Still useful when they fit:
python3 tools/dfg_inspect.py <file> driver <name>python3 tools/dfg_inspect.py <file> cone <name> [--depth N]python3 tools/dfg_inspect.py <file> mux <name>python3 tools/dfg_inspect.py <file> mux-case <id-or-name> <case_val>python3 tools/dfg_inspect.py <file> const_driven [value]python3 tools/dfg_inspect.py <file> group <prefix>python3 tools/dfg_inspect.py <file> search <pattern>When debugging:
node --details for the suspicious signal or node.deps to walk backward through the inputs that determine the value.uses to see downstream impact and to find the first consumer that changes across passes.neighborhood when the issue is local and you want one compact view.nodes --op OP or nodes --at FILE:LINE when you know the operation kind or source location but not the exact name.diff on adjacent passes first. Add --node or --cone once the suspect area is known.trace-dfg-runtime.Notes:
instance_path, full_name, and stable debug_id values. Prefer those when present.