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/babyworm/rtl-agent-team --skill cdc-tool-profiles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。
正在显示 SKILL.md
| name | cdc-tool-profiles |
| description | Internal reference: cdc tool profiles (agent-loaded; do not invoke). |
| user-invocable | false |
svlens provides quantitative measurements (scores, violation counts, complexity metrics) to add consistency to the LLM's qualitative judgment. Phase gate decisions require BOTH:
Neither alone is sufficient. A zero-violation svlens report does not auto-pass a gate if the LLM identifies architectural concerns. Conversely, the LLM cannot override quantitative violations without documented justification (waiver with rationale).
When commercial tools are available, they serve as the signoff authority. svlens provides supplementary crosscheck data — useful for early-stage CI, pre-signoff screening, and adding quantitative rigor to LLM-driven reviews.
lint/scripts/run_cdc.shVIOLATIONCAUTIONCONVENTIONINFOWAIVEDVIOLATION exists.svlens is a unified structural analysis toolkit with three modes sharing one elaboration:
lint/scripts/run_cdc.sh --tool structural --top <top> -f rtl/filelist_top.f --outdir lint/cdc
(structural mode auto-runs svlens crosscheck when installed)lint/scripts/run_cdc.sh --tool svlens --top <top> -f rtl/filelist_top.f --outdir lint/cdc
(standalone mode)summary.violations, summary.cautions, per-crossing severity + sync_typesvlens conn --format all --top <top> -f rtl/filelist_top.f -o sim/conn/reports--check-protocol), naming convention (--check-convention)--expect connectivity_spec.yaml)analysis.overall_score, analysis.module_health, summary.errorssvlens metrics --format all --top <top> -f rtl/filelist_top.f -o sim/metrics/reportslogic_depth_est, raw_node_count, source_inputslogic_depth_est > 15-20 = timing risk, source_inputs > 20 = complex convergencesvlens all --format all --top <top> -f rtl/filelist_top.f -o sim/svlens/reportsconn/, cdc/, metrics/svlens_summary.json with aggregated results--diff baseline/ for regression detectiongit clone https://github.com/babyworm/svlens.git ~/tools/svlens
cd ~/tools/svlens
./scripts/setup-deps.sh --prefix ~/.local
cmake -B build -DCMAKE_PREFIX_PATH=~/.local
cmake --build build -j$(nproc)
cmake --install build --prefix ~/.local
spyglass (binary: sg_shell, config key: sg_shell, script flag: --tool spyglass):
lint/scripts/run_cdc.sh --tool spyglass --top <top> -f rtl/filelist_top.f --outdir lint/cdcrat_config.json uses sg_shell as tool key; runner scripts accept --tool spyglassvc_cdc (binary: vc_static_shell, script flag: --tool vc_cdc):
lint/scripts/run_cdc.sh --tool vc_cdc --top <top> -f rtl/filelist_top.f --outdir lint/cdcvc_cdc is the Synopsys app name; VC SpyGlass CDC runs inside the VC
Static shell, so the executable is vc_static_shellquesta_cdc:
lint/scripts/run_cdc.sh --tool questa_cdc --top <top> -f rtl/filelist_top.f --outdir lint/cdcThe exact binaries and batch switches the runner uses. These are the shapes each vendor documents; changing them silently breaks the stage.
| Tool | Invocation | Why this shape |
|---|---|---|
structural / svlens | svlens cdc --format all -o <outdir> | — |
spyglass | sg_shell -tcl <tcl> | — |
vc_cdc | vc_static_shell -f <tcl> -batch | vc_cdc is an app name, not an executable. -batch makes the shell quit on an unexpected error instead of dropping to a prompt |
questa_cdc | qverify -c -do <do> -od <outdir> | -c selects CLI/batch. Without -od the session database lands in the default location while the runner reports under --outdir |
toolviolations_countcautions_countinfo_countreport_pathreplay_path