菜单
Review synthesizable SystemVerilog for static RTL issues such as inferred latches, reset mismatches, width hazards, and multi-driver structure. Use this skill whenever generated RTL needs a deterministic front-end lint audit before CDC, RDC, or timing review.
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。
基于 SOC 职业分类
Extract a source-grounded, AI-readable textual design view from HDL source, parser output, MCP tool output, or model reasoning. Use this skill when the user needs Verilog, SystemVerilog, VHDL, or proprietary hardware description language normalized into a reusable design view for downstream analysis flows such as timing, lint, CDC, RDC, formal planning, or DV planning. Prefer UHDM text for SystemVerilog/Verilog when available, AST JSON when UHDM is not available, and explicit model-derived views only when no tool evidence exists.
Analyze an HDL design view or visible RTL for pre-synthesis timing risk by estimating combinational logic depth on register-to-register and boundary timing paths. Use this skill when the user already has a UHDM text dump, AST JSON, source-grounded textual design view, or small visible RTL block and asks for critical paths, deep combinational logic, reg-to-reg timing risk, or pre-synthesis timing feedback. This skill is analysis-only; use an HDL-design-view extraction skill or flow when parser/tool/MCP selection, source-language normalization, or design-view generation is required.
Read an approved microarchitecture specification and emit synthesizable SystemVerilog with requirement traceability. Use this skill whenever the flow moves from block architecture into concrete RTL implementation.
Normalize a block-level design brief into deterministic requirements, explicit PPA targets, and open intake questions. Use this skill whenever the user provides high-level microarchitectural intent and the workflow must capture power, performance, and area before specification or RTL work begins.
Assemble normalized requirements, the microarchitecture spec, generated RTL, and static audit summaries into one front-end handoff package. Use this skill whenever the block-level RTL planning flow needs one deterministic output that can feed downstream DV planning.
Turn normalized block requirements into a Markdown microarchitecture specification with requirement traceability and targeted diagrams. Use this skill whenever the workflow needs a deterministic architecture document before RTL generation.
| name | rtl-lint-auditor |
| description | Review synthesizable SystemVerilog for static RTL issues such as inferred latches, reset mismatches, width hazards, and multi-driver structure. Use this skill whenever generated RTL needs a deterministic front-end lint audit before CDC, RDC, or timing review. |
Audit block-level SystemVerilog for front-end lint issues and emit one deterministic finding report.
Scope: block or small IP. This is a static lint audit, not an EDA signoff run.
Apply these repo rules before analysis and output generation:
../../rules/common/evidence-grounding.md../../rules/common/output-discipline.md../../rules/common/tool-evidence-provenance.md../../rules/rtl/synthesizable-systemverilog.md../../rules/rtl/lint-severity.mdRead the RTL and perform these steps:
project_root, source files, or a filelist are provided and HDL/EDA MCP
tools are available, collect read-only compiler or lint evidence before
finalizing findings.../../rules/rtl/lint-severity.md.critical and high findings as blocking.findings list when the RTL is clean under this rule set.When MCP tools are available, use them as evidence sources before finalizing findings. Discover tools by schema and description, not by vendor name. Prefer read-only HDL/EDA tools that provide parse/elaboration diagnostics, design-unit inventory, hierarchy, symbol lookup, or lint-rule-deck reports.
pyslang-mcp can ground front-end diagnostics, module descriptions, hierarchy,
and symbol references. Useful calls include pyslang_parse_filelist or
pyslang_parse_files, pyslang_get_diagnostics,
pyslang_list_design_units, pyslang_describe_design_unit,
pyslang_get_hierarchy, and pyslang_find_symbol. Treat
pyslang-mcp as compiler-backed context, not as a replacement for a full lint
deck.
For vendor or internal EDA MCPs, use tools only when their schema clearly exposes read-only parse/elaboration diagnostics, lint report retrieval, design-unit or hierarchy queries, symbol lookup, or waiver/status lookup that is explicitly in scope. Do not assume public vendor MCP support or hard-code tool names.
Normalize every tool finding into this skill's deterministic YAML severity model.
Do not blindly copy vendor severities when they conflict with
../../rules/rtl/lint-severity.md; preserve original tool rule IDs, severities,
or diagnostics only as evidence metadata.
Do not claim lint cleanliness solely because pyslang-mcp returns zero
diagnostics. That means the design parsed under that frontend; text-grounded lint
hazards can still remain.
If MCP evidence is unavailable, incomplete, or returns a structured tool error,
continue with text-grounded analysis and record the limitation in
tool_evidence when a tool path was attempted. Emit tool_evidence: [] when no
MCP evidence was used. When tool evidence is present, each entry must contain
source, tools, purpose, status, and summary. Do not invoke destructive,
synthesis, simulation, CDC, timing, or RTL-editing tools from this skill.
Produce a single YAML document:
design:
top_module: event_counter
rtl_files:
- rtl/event_counter.sv
findings:
- id: LINT-001
severity: high
category: latch
file: rtl/event_counter.sv
line: 27
message: "The combinational next-state block leaves irq_nxt unassigned on one branch."
recommendation: "Assign irq_nxt on every path or provide a default assignment before the branch tree."
blocking: true
tool_evidence: []
summary:
total_findings: 1
blocking_findings: 1
by_severity:
critical: 0
high: 1
medium: 0
low: 0
info: 0