菜单
Audit a Claude Code skill file against the official skills specification. Use when the user asks to review, audit, or check a skill for spec compliance, or when a skill file may be too long, have wrong frontmatter, or needs structural improvement.
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
Import P&R (Genus + Innovus) products into Virtuoso: GDS layout, Verilog schematic/symbol, power labels, and label restyling. Four-step pipeline driven entirely from vcli skill exec — no Python bridge or GUI required. Use when: (1) user wants to pull a routed GDS or post-P&R netlist into Virtuoso, (2) layout labels look giant/unreadable after import, (3) user mentions strmin / ihdl / digital import / P&R-to-schematic flow.
Generate Virtuoso schematics from topology descriptions via vcli schematic commands. Use when: (1) user wants to draw/create a schematic in Virtuoso, (2) user says "draw the OTA" or "create the schematic", (3) after sizing is complete and ready to build the circuit, (4) user provides a topology and wants it instantiated.
Read Maestro/ADE simulation output values directly from PSF binary files — no Virtuoso GUI or bridge required. Parses maestro.sdb + active.state XML to extract output expressions, resolves the PSF directory from history.sdb, evaluates Ocean-style expressions (getData, dB20, phaseDeg, bandwidth, ymax, VF, VT…), and returns structured JSON results. Use this skill whenever: - The user asks "what is the gain / phase margin / bandwidth from the simulation?" - `maeGetOutputValue` returned nil (requires GUI results loaded in memory) - The user wants to check simulation results offline or from a script - The user wants to read PSF files / evaluate output expressions from a Maestro session - Any request involving Maestro output expressions, PSF files, or ADE results without GUI
Maestro (ADE Assembler) session management and simulation. Use when: running simulations via Maestro, configuring tests/analyses/outputs, updating design variables, reading results.
Evolve skills from vcli session history. Use when: (1) user says "根據歷史進化技能" / "evolve skills from history" / "覆盤技能", (2) after a debugging session to capture what went wrong, (3) periodic skill maintenance to close knowledge gaps revealed by real usage. Reads cmd.jsonl + per-session SKILL logs, finds failure/correction/gap signals, maps them to skills in .claude/skills/, and writes concrete improvements.
Execute SKILL code on Virtuoso. Use when running SKILL expressions, querying cellview data, listing libraries/cells, or interacting with Virtuoso programmatically.
| name | skill-review |
| description | Audit a Claude Code skill file against the official skills specification. Use when the user asks to review, audit, or check a skill for spec compliance, or when a skill file may be too long, have wrong frontmatter, or needs structural improvement. |
| argument-hint | ["skill name or path","e.g. \"veriloga\" or \".claude/skills/veriloga/SKILL.md\""] |
| allowed-tools | Read Bash(wc *) Bash(find *) Bash(ls *) |
Audit the skill named "$ARGUMENTS". If no argument was given, ask the user which skill to review.
find .claude/skills ~/.claude/skills -name "SKILL.md" 2>/dev/null | head -30
wc -l <path/to/SKILL.md>
ls -la <skill-dir>/
Read the SKILL.md file in full.
| Check | Limit | Notes |
|---|---|---|
| SKILL.md line count | ≤ 500 lines | Excess is silently truncated before Claude sees it |
| reference.md exists when needed | — | Required if SKILL.md would overflow |
Module templates, long examples, and reference tables belong in reference.md, not the main file.
| Field | Required | Constraint |
|---|---|---|
name | Yes | Must match the slash-command name exactly |
description | Yes | ≤ 1,536 chars; drives auto-trigger matching |
allowed-tools | Recommended | ToolName or ToolName(glob); glob uses bare *, not */prefix/ |
argument-hint | Recommended | Short hint shown in autocomplete UI |
allowed-tools patterns:
CORRECT: Bash(virtuoso *) Bash(vcli *) Read Write
WRONG: Bash(*/virtuoso *) — leading */ has unclear glob semantics
WRONG: Bash — no glob = allows ALL bash commands (too broad)
| Check | What to look for |
|---|---|
| Arguments routing | Body references the ARGUMENTS variable to route on user input |
| Skill-dir-relative paths | Companion files use the CLAUDE_SKILL_DIR variable, not hardcoded absolute paths |
| Dynamic blocks | Time-sensitive content (git branch, date, env vars) injected at load time |
| No hardcoded paths | Absolute paths to project dirs should not appear in skill body |
| Check | Guidance |
|---|---|
| Templates vs reference.md | Multiple long code blocks (>30 lines each) → extract to reference.md |
| No commented-out code | Clean, actionable content only |
| Runnable examples | Snippets work as-is or with minimal substitution |
Produce a findings table:
| # | Check | Status | Finding |
|---|-------------------|--------|---------------------------------------------------|
| 1 | File size | FAIL | 792 lines — move templates to reference.md |
| 2 | allowed-tools | WARN | Bash(*/vcli *) — remove leading */ |
| 3 | argument-hint | FAIL | Missing from frontmatter |
| 4 | Arguments routing | WARN | ARGUMENTS variable not used — skill ignores input |
List recommended fixes in priority order (FAIL first, then WARN).
Ask: "Apply all fixes now?" If yes:
reference.md in the same directoryallowed-tools, argument-hint, namereference.mddescription is used for semantic matching — keep it focused, under ~1,000 charsallowed-tools controls which tools Claude may call during skill execution