skill-advisor
npx skills add https://github.com/v5tech/skills --skill skill-advisor
Agent 安装分布
Skill 文档
Skill Advisor
Help users search, compare, and install skills correctly to their preferred agents.
Output Language: Match your response language to the user’s conversation language. If user writes in Chinese, respond in Chinese; if in English, respond in English.
Default Target Agents
Install skills globally to these 4 agents by default:
| Agent | –agent value |
|---|---|
| Antigravity | antigravity |
| Claude Code | claude-code |
| Gemini CLI | gemini-cli |
| OpenCode | opencode |
User can customize this list by specifying different agents.
Workflow
Step 1: Search Skills
Search with multiple related keywords for better coverage:
npx skills find [query]
npx skills find [related-query]
Step 2: Preview Skill Content
Before comparing, preview each repository’s available skills:
npx skills add <owner/repo> -g --list
For promising skills, read the actual SKILL.md to understand real functionality. Visit:
https://skills.sh/<owner>/<repo>/<skill>
# Example: https://skills.sh/expo/skills/building-native-ui
Never rely solely on skill names to judge functionality. Names can be misleading.
Step 3: Compare and Recommend
Evaluate each skill on these dimensions:
| Dimension | How to Assess | Weight |
|---|---|---|
| Content quality | Read SKILL.md â does it contain actionable guidance, scripts, or references? Or is it generic/empty? | High |
| Coverage systematics | Are multiple skills from the same author forming a coherent system? | High |
| Tech stack match | Does the skill match the user’s current project tech stack? | High |
| Source authority | Official (e.g., expo/skills) > Well-known org > Individual | Medium |
| Maintenance activity | Repository update date, commit frequency | Medium |
| Depth vs breadth | Deep expertise in one area vs shallow generic guidance | Medium |
| Function overlap | Does it duplicate functionality of another found skill? | Medium |
| Dependencies | Do multiple skills need to work together? | Low |
Comparison output format:
## Recommended (Category/Author)
| Skill | Function | Quality | Note |
|-------|----------|---------|------|
| skill-a | Actual function from SKILL.md | High/Medium/Low | Why recommended |
## Optional
| Skill | Function | Quality | Note |
|-------|----------|---------|------|
| skill-b | Actual function from SKILL.md | Medium | Supplements skill-a |
## Not Recommended
| Skill | Function | Quality | Note |
|-------|----------|---------|------|
| skill-c | Actual function from SKILL.md | Low | Overlaps with skill-a, lower quality |
Quality assessment criteria:
- High: Contains scripts, detailed references, concrete examples, actionable workflows
- Medium: Has useful guidance but mostly text-based instructions
- Low: Generic/vague content, could be replaced by general LLM knowledge
Step 4: Generate Install Commands
CRITICAL: Never use --all flag â it overrides --agent and installs to ALL agents.
Install command generation rules:
- Only generate install commands for High quality skills
- Medium quality skills are listed in “Optional” section, commands provided separately for user decision
- Low quality skills in “Not Recommended” section get no install commands
Group by repository: If multiple High quality skills come from the same repository, group them in one command.
Correct command pattern:
npx skills add <owner/repo> -g -s <skill1> -s <skill2> -a antigravity -a claude-code -a gemini-cli -a opencode -y
Parameter reference:
| Parameter | Meaning |
|---|---|
-g, --global |
Global install (user-level) |
-s, --skill <skills> |
Specify skill names (multiple values accepted, use '*' for all) |
-a, --agent <agents> |
Target agents (multiple values accepted, use '*' for all) |
-y, --yes |
Skip confirmation |
-l, --list |
List available skills without installing |
--all |
Install all skills to all agents without prompts |
Wrong vs Right:
# WRONG â --all overrides -a, installs to ALL 39+ agents
npx skills add expo/skills -g -a antigravity -a claude-code -a gemini-cli -a opencode --all
# RIGHT â repeat -s for each skill, -a for each agent
npx skills add expo/skills -g -s building-native-ui -s expo-deployment -a antigravity -a claude-code -a gemini-cli -a opencode -y
Step 5: Verify Installation
# List installed skills and their agents
npx skills ls -g
# Check symlinks are valid in each agent directory
ls -la ~/.claude/skills/
ls -la ~/.gemini/skills/
ls -la ~/.gemini/antigravity/skills/
ls -la ~/.config/opencode/skills/
# Verify no extra agent directories created
ls ~/.cursor/ ~/.cline/ ~/.copilot/ 2>/dev/null
Recovery
Remove from Specific Agent
Goal: Remove skill from one agent while keeping it available for other agents
Operation: Only delete the symlink, preserve source files
# Remove from Claude Code only
rm ~/.claude/skills/<skill-name>
# Remove from Gemini CLI only
rm ~/.gemini/skills/<skill-name>
# Remove from Antigravity only
rm ~/.gemini/antigravity/skills/<skill-name>
# Remove from OpenCode only
rm ~/.config/opencode/skills/<skill-name>
Source files in ~/.agents/skills/ remain intact. Other agents can still use this skill.
Remove from All Agents
Goal: Fully remove skill, delete source files, clean all symlinks
Operation: Use CLI to delete source files, then manually clean up any orphaned symlinks
# Step 1: CLI removes source files + lock records + detected symlinks
npx skills remove -g -s <skill-name> -y
# Step 2: Clean up any remaining orphaned symlinks (CLI may miss some agents)
for dir in ~/.claude/skills ~/.gemini/skills ~/.gemini/antigravity/skills ~/.config/opencode/skills; do
[ -L "$dir/<skill-name>" ] && rm "$dir/<skill-name>" && echo "Cleaned: $dir/<skill-name>"
done
Reinstall to Specific Agents
Goal: Clean removal then reinstall to specified agents
Operation: Use CLI to delete source files, clean symlinks for target agents, then reinstall
# Step 1: CLI removes source files + lock records
npx skills remove -g -s <skill-name> -y
# Step 2: Clean up symlinks for target agents
for dir in ~/.claude/skills ~/.gemini/skills ~/.gemini/antigravity/skills ~/.config/opencode/skills; do
[ -L "$dir/<skill-name>" ] && rm "$dir/<skill-name>" && echo "Cleaned: $dir/<skill-name>"
done
# Step 3: Reinstall to specified agents
npx skills add <owner/repo> -g -s <skill-name> -a antigravity -a claude-code -a gemini-cli -a opencode -y
Batch Cleanup
When installation went wrong (e.g., used --all by accident):
1. Remove skills source files
# Remove specific skill (CLI handles lock file)
npx skills remove -g -s <skill-name> -y
# Or remove multiple skills
npx skills remove -g -s <skill1> -s <skill2> -y
2. Clean up all symlinks
for dir in ~/.claude/skills ~/.gemini/skills ~/.gemini/antigravity/skills ~/.config/opencode/skills; do
[ -d "$dir" ] && find "$dir" -maxdepth 1 -type l -delete -print
done
3. Clean empty agent directories
Check each non-target agent directory. Only delete if:
- Directory was newly created by skills CLI
- Contains only an empty
skills/subdirectory - Has no other files
# Check before deleting
ls -la ~/.cursor/
find ~/.cursor/ -type f | wc -l # Must be 0 to safely delete
4. Reinstall correctly
npx skills add <owner/repo> -g -s <skill1> -s <skill2> -a antigravity -a claude-code -a gemini-cli -a opencode -y
Update Skills
npx skills check # Check for available updates
npx skills update # Update all installed skills
Agent Directory Reference
See references/agents.md for the complete list of all 39 agents and their global paths.
| Agent | Global Path |
|---|---|
| Antigravity | ~/.gemini/antigravity/skills/ |
| Claude Code | ~/.claude/skills/ |
| Gemini CLI | ~/.gemini/skills/ |
| OpenCode | ~/.config/opencode/skills/ |
Common Pitfalls
--allflag: Overrides--agent, installs to 39+ agents, creates unwanted directories- Forgetting
-g: Without global flag, installs to project only - Missing
-s: Without specifying skills, may install all or prompt interactively - Trusting skill names: Always read SKILL.md content before recommending