skill-advisor

这段内容是审核时保存的快照。外部来源才是完整且最新的版本。

---
name: skill-advisor
description: Search, compare, and install skills via npx skills CLI. Use when user wants to find skills, compare functionality, or install to agents. Avoids --all pitfall, targets Antigravity, Claude Code, Gemini CLI, OpenCode by default.
---

# Skill Advisor

Search, compare, and install skills correctly to preferred agents.

## Determine Operation Type

**What does the user want?**

- **Find and install new skills** → Follow [Installation Workflow](#installation-workflow)
- **Remove existing skills** → Read [references/recovery.md](references/recovery.md)
- **Fix broken installation** → Read [references/recovery.md](references/recovery.md)
- **Update skills** → Run `npx skills check` then `npx skills update`

## Installation Workflow

Copy this checklist and track progress:

```
Skill Installation:
- [ ] Step 1: Search with multiple keywords
- [ ] Step 2: Preview and read SKILL.md
- [ ] Step 3: Compare and categorize (High/Medium/Low)
- [ ] Step 4: Generate install commands (no --all!)
- [ ] Step 5: Verify installation
```

### Step 1: Search Skills

Search with multiple related keywords for better coverage:

```bash
npx skills find QUERY
npx skills find RELATED_QUERY
```

### Step 2: Preview and Read SKILL.md

Preview each repository's available skills:

```bash
npx skills add owner/repo -g --list
```

Read actual SKILL.md content at:
```
https://skills.sh/owner/repo/skill-name
```

**Never rely solely on skill names.** Names can be misleading.

### Step 3: Compare and Recommend

Evaluate each skill:

| Dimension | Weight |
|-----------|--------|
| Content quality (scripts, references, actionable workflows) | High |
| Coverage systematics (coherent skill system from same author) | High |
| Tech stack match | High |
| Source authority (Official > Org > Individual) | Medium |
| Maintenance activity | Medium |
| Function overlap | Medium |

**Quality criteria:**
- **High**: Scripts, detailed references, concrete examples, actionable workflows
- **Medium**: Useful guidance but mostly text-based
- **Low**: Generic content, replaceable by general LLM knowledge

**Output format:**

```markdown
## Recommended (Category/Author)
| Skill | Function | Quality | Note |
|-------|----------|---------|------|
| skill-a | Actual function from SKILL.md | High | Why recommended |

## Optional
| Skill | Function | Quality | Note |
|-------|----------|---------|------|
| skill-b | Function from SKILL.md | Medium | Supplements skill-a |

## Not Recommended
| Skill | Function | Quality | Note |
|-------|----------|---------|------|
| skill-c | Function from SKILL.md | Low | Overlaps with skill-a |
```

### Step 4: Generate Install Commands

**CRITICAL: Never use `--all` flag** — it overrides `--agent` and installs to ALL 41 agents.

**Rules:**
1. Only generate install commands for **High** quality skills
2. **Medium** quality: list in Optional section, commands provided separately
3. **Low** quality: no install commands

**Correct pattern:**

```bash
npx skills add owner/repo -g -s skill1 -s skill2 -a antigravity -a claude-code -a gemini-cli -a opencode -y
```

**Parameters:**

| Flag | Meaning |
|------|---------|
| `-g` | Global install (user-level) |
| `-s SKILL` | Specify skill (repeat for multiple) |
| `-a AGENT` | Target agent (repeat for multiple) |
| `-y` | Skip confirmation |
| `-l` | List available skills only |
| `--all` | **DANGER**: Installs all skills to all agents |

**Wrong vs Right:**

```bash
# WRONG — --all overrides -a, installs to ALL agents
npx skills add expo/skills -g -a claude-code --all

# RIGHT — explicit -s and -a flags
npx skills add expo/skills -g -s xlsx -s pdf -a antigravity -a claude-code -a gemini-cli -a opencode -y
```

### Step 5: Verify Installation

```bash
# List installed skills
npx skills ls -g

# Check symlinks exist
ls -la ~/.claude/skills/
ls -la ~/.gemini/skills/
ls -la ~/.gemini/antigravity/skills/
ls -la ~/.config/opencode/skills/

# Verify no unwanted directories
ls ~/.cursor