The docs site had drifted (345 skills / 80 plugins against 411 / 110,
~150 pages missing from the nav, 32 pages with install commands for
plugins that do not exist). The fix is to stop hand-maintaining it.
- MkDocs hook (scripts/docs_hooks.py) runs the generator, builds the
Skills/Agents/Commands nav from what it generated, and fills
{{ count.* }}, {{ plugin_table }} and {{ whats_new }} tokens in
hand-written pages and llms.txt
- generate-docs.py: install lines resolve to real marketplace plugins,
c-level-agents and loop-library domains added, vendored links mapped
to the real files, edit buttons point at the source SKILL.md, writes
only when content changes
- Generated skill/agent/command pages are no longer committed; 42
redirects keep moved URLs working
- New What's New page (from CHANGELOG.md), Hermes and Vibe guides,
support matrix; Vibe path corrected to the flat ~/.vibe/skills layout
- Fixed broken install instructions (OpenClaw curl one-liner, claude
/plugin form, nonexistent plugins) in guides and README; README
install order now starts with Claude Code
- Pinned mkdocs versions; Pages build runs --strict (0 warnings)
- Gate G14 (scripts/check_docs_site.py): hard-coded counters, unknown
install targets and nav orphans fail CI
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01L8KYFH1Jfe2JqY8uFr4aBx
29 KiB
| title | description |
|---|---|
| Cursor, Aider, Windsurf, Hermes, Mistral Vibe & 8 More AI Coding Tools | Install Claude Code skills and agent plugins in Hermes Agent, Mistral Vibe, Cursor, Aider, Kilo Code, Windsurf, OpenCode, Augment, and Antigravity. One-command conversion for 13 AI coding agents. |
Multi-Tool Integrations
All {{ count.skills }} skills in this repository work with 9 AI coding tools beyond Claude Code, Codex, Gemini CLI, and OpenClaw. Hermes Agent and Mistral Vibe both use the same agentskills.io SKILL.md standard — no conversion needed. For the other 7 tools, a conversion script adapts the format each tool expects while preserving skill instructions, workflows, and supporting files.
Support matrix
How much each tool is tested tells you how much to trust it. "CI install-tested" means every pull request that changes skills installs the real tool into an empty home directory and fails unless the tool itself lists every installed skill.
| Tool | Format | How you install | Tested |
|---|---|---|---|
| Claude Code | Native plugins | /plugin install <name>@claude-code-skills |
Clean install of every plugin before each release |
| OpenAI Codex | Native SKILL.md |
./scripts/codex-install.sh |
CI install-tested ({{ count.codex }} skills) |
| Gemini CLI | Native SKILL.md |
./scripts/gemini-install.sh |
CI install-tested ({{ count.gemini }} skills) |
| Hermes Agent | Native SKILL.md |
python3 scripts/sync-hermes-skills.py |
CI install-tested ({{ count.hermes }} skills) |
| Mistral Vibe | Native SKILL.md |
./scripts/vibe-install.sh |
CI install-tested ({{ count.vibe }} skills) |
| OpenClaw | Native SKILL.md |
./scripts/openclaw-install.sh |
Not tested in CI |
| Cursor, Aider, Kilo Code, Windsurf, OpenCode, Augment, Antigravity | Converted | ./scripts/convert.sh then ./scripts/install.sh --tool <name> |
Conversion only, not tested in the tool |
-
:material-cursor-default-click:{ .lg .middle } Cursor
.mdcrule files in.cursor/rules/ -
:material-code-braces:{ .lg .middle } Aider
Single
CONVENTIONS.mdfile -
:material-alpha-k-box:{ .lg .middle } Kilo Code
Markdown rules in
.kilocode/rules/ -
:material-surfing:{ .lg .middle } Windsurf
SKILL.mdbundles in.windsurf/skills/ -
:material-console:{ .lg .middle } OpenCode
SKILL.mdbundles in.opencode/skills/ -
:material-auto-fix:{ .lg .middle } Augment
Rule files in
.augment/rules/ -
:material-google:{ .lg .middle } Antigravity
SKILL.mdbundles in~/.gemini/antigravity/skills/ -
:material-medical-bag:{ .lg .middle } Hermes Agent
Native
SKILL.mdin~/.hermes/skills/— no conversion needed -
:material-wave:{ .lg .middle } Mistral Vibe
Native
SKILL.mdin~/.vibe/skills/— no conversion needed
Quick Start
1. Convert
git clone https://github.com/alirezarezvani/claude-skills.git
cd claude-skills
# Convert all skills for all tools (~15 seconds)
./scripts/convert.sh --tool all
# Or convert for a specific tool only
./scripts/convert.sh --tool cursor
2. Install
# Install into your project directory
./scripts/install.sh --tool cursor --target /path/to/project
# Or install globally (Antigravity)
./scripts/install.sh --tool antigravity
# Skip confirmation prompts
./scripts/install.sh --tool aider --target . --force
3. Verify
Each tool section below includes a verification step to confirm skills are loaded.
!!! tip "Regenerate after updates"
When you pull new skills from the repository, re-run ./scripts/convert.sh and ./scripts/install.sh to update your local installation.
How Conversion Works
The converter reads each skill's SKILL.md frontmatter (name and description) and markdown body, then outputs the format each tool expects:
| Source | Target | What Changes |
|---|---|---|
| YAML frontmatter | Tool-specific frontmatter | Field names/values adapted per tool |
| Markdown body | Passed through | Instructions preserved as-is |
scripts/ dir |
Copied (where supported) | Antigravity, Windsurf, OpenCode |
references/ dir |
Copied (where supported) | Antigravity, Windsurf, OpenCode |
templates/ dir |
Copied (where supported) | Antigravity, Windsurf, OpenCode |
Tools that use flat files (Cursor, Aider, Kilo Code, Augment) get the SKILL.md body only — supporting directories are not copied since those tools don't support subdirectories per rule.
Cursor
Cursor uses .mdc rule files in .cursor/rules/ with frontmatter for description, glob patterns, and auto-apply settings.
Format
Each skill becomes a single .mdc file:
---
description: "What this skill does and when to activate it"
globs:
alwaysApply: false
---
# Skill instructions here...
alwaysApply: false— skills are available on-demand, not always loadedglobs:— empty by default; add file patterns to auto-activate for specific files (e.g.,*.test.ts)
Install
=== "Script"
```bash
./scripts/convert.sh --tool cursor
./scripts/install.sh --tool cursor --target /path/to/project
```
=== "Manual"
```bash
mkdir -p /path/to/project/.cursor/rules
cp integrations/cursor/rules/*.mdc /path/to/project/.cursor/rules/
```
Verify
find .cursor/rules -name "*.mdc" | wc -l
# Expected: 156
Open the Cursor rules panel to see all available skills listed.
Customization
After installation, you can:
- Set
alwaysApply: trueon skills you want active in every conversation - Add
globs: "*.py"to auto-activate Python-related skills for.pyfiles - Remove skills you don't need to keep your rules panel clean
Aider
Aider reads a CONVENTIONS.md file from your project root. All skills are concatenated into this single file with section headers.
Format
# Claude Skills — Aider Conventions
> Auto-generated from claude-skills. Do not edit manually.
> Generated: 2026-03-11
---
## copywriting
> When the user wants to write, rewrite, or improve marketing copy...
# Copywriting
You are an expert conversion copywriter...
---
## senior-architect
> Deep expertise in system architecture...
# Senior Architect
...
Install
=== "Script"
```bash
./scripts/convert.sh --tool aider
./scripts/install.sh --tool aider --target /path/to/project
```
=== "Manual"
```bash
cp integrations/aider/CONVENTIONS.md /path/to/project/
```
Usage
# Aider automatically reads CONVENTIONS.md from the project root
aider
# Or explicitly point to it
aider --read CONVENTIONS.md
Verify
wc -l CONVENTIONS.md
# Expected: one rule file per skill ({{ count.skills }} skills)
grep -c "^## " CONVENTIONS.md
# Expected: 156 (one section per skill)
!!! note "Large file"
The combined CONVENTIONS.md is ~41K lines. Aider handles this well, but if you prefer a smaller file, you can edit it to keep only the skills relevant to your project.
Kilo Code
Kilo Code reads plain markdown rules from .kilocode/rules/. No special frontmatter required.
Format
Each skill becomes a clean markdown file:
# copywriting
> When the user wants to write, rewrite, or improve marketing copy...
# Copywriting
You are an expert conversion copywriter...
Install
=== "Script"
```bash
./scripts/convert.sh --tool kilocode
./scripts/install.sh --tool kilocode --target /path/to/project
```
=== "Manual"
```bash
mkdir -p /path/to/project/.kilocode/rules
cp integrations/kilocode/rules/*.md /path/to/project/.kilocode/rules/
```
Verify
find .kilocode/rules -name "*.md" | wc -l
# Expected: 156
Open Kilo Code's rules panel (click the ⚖ icon) to see all rules loaded.
Mode-Specific Rules
Kilo Code supports mode-specific rules. To assign skills to specific modes:
# Move architecture skills to "architect" mode
mkdir -p .kilocode/rules-architect/
mv .kilocode/rules/senior-architect.md .kilocode/rules-architect/
mv .kilocode/rules/database-designer.md .kilocode/rules-architect/
Windsurf
Windsurf uses the same SKILL.md format as Claude Code — skills convert with minimal changes.
Format
Each skill becomes a directory with SKILL.md plus optional supporting files:
.windsurf/skills/copywriting/
├── SKILL.md # Instructions with name/description frontmatter
├── scripts/ # Python tools (if present in source)
├── references/ # Domain knowledge (if present)
└── templates/ # Code templates (if present)
---
name: "copywriting"
description: "When the user wants to write, rewrite, or improve marketing copy..."
---
# Copywriting
...
Install
=== "Script"
```bash
./scripts/convert.sh --tool windsurf
./scripts/install.sh --tool windsurf --target /path/to/project
```
=== "Manual"
```bash
cp -R integrations/windsurf/skills/* /path/to/project/.windsurf/skills/
```
Verify
find .windsurf/skills -name "SKILL.md" | wc -l
# Expected: 156
Skills appear automatically in Windsurf's skill list. You can also invoke them with @skill-name.
Progressive Disclosure
Windsurf uses progressive disclosure — only the skill name and description are shown by default. The full SKILL.md content loads only when Windsurf decides the skill is relevant to your request, keeping your context window lean.
OpenCode
OpenCode supports skills in .opencode/skills/ with SKILL.md files. It also reads Claude Code's .claude/skills/ as a fallback.
Format
Each skill becomes a directory with SKILL.md:
---
name: "copywriting"
description: "When the user wants to write, rewrite, or improve marketing copy..."
compatibility: opencode
---
# Copywriting
...
The compatibility: opencode field is added to help OpenCode identify these as native skills.
Install
=== "Script"
```bash
./scripts/convert.sh --tool opencode
./scripts/install.sh --tool opencode --target /path/to/project
```
=== "Manual"
```bash
cp -R integrations/opencode/skills/* /path/to/project/.opencode/skills/
```
=== "Global"
```bash
# Install globally for all projects
cp -R integrations/opencode/skills/* ~/.config/opencode/skills/
```
Verify
find .opencode/skills -name "SKILL.md" | wc -l
# Expected: 156
Claude Code Compatibility
OpenCode also reads .claude/skills/ directories. If you already have skills installed for Claude Code, OpenCode will discover them automatically — no conversion needed.
To disable this fallback:
export OPENCODE_DISABLE_CLAUDE_CODE_SKILLS=1
Augment
Augment reads rule files from .augment/rules/ with frontmatter specifying activation type.
Format
Each skill becomes a markdown rule file:
---
type: auto
description: "When the user wants to write, rewrite, or improve marketing copy..."
---
# Copywriting
...
type: auto— Augment automatically activates the rule when it matches your request- Other types:
always(always loaded),manual(user-invoked only)
Install
=== "Script"
```bash
./scripts/convert.sh --tool augment
./scripts/install.sh --tool augment --target /path/to/project
```
=== "Manual"
```bash
mkdir -p /path/to/project/.augment/rules
cp integrations/augment/rules/*.md /path/to/project/.augment/rules/
```
Verify
find .augment/rules -name "*.md" | wc -l
# Expected: 156
Customization
Change type: auto to type: always for skills you want loaded in every conversation:
# Make coding standards always active
sed -i 's/type: auto/type: always/' .augment/rules/senior-architect.md
Antigravity
Antigravity (Google) uses SKILL.md files in ~/.gemini/antigravity/skills/ with additional metadata fields.
Format
---
name: "copywriting"
description: "When the user wants to write, rewrite, or improve marketing copy..."
risk: low
source: community
date_added: '2026-03-11'
---
# Copywriting
...
Additional fields:
risk: low— all skills are instruction-only, no dangerous operationssource: community— identifies these as community-contributed skillsdate_added— conversion date for tracking freshness
Install
=== "Script"
```bash
./scripts/convert.sh --tool antigravity
./scripts/install.sh --tool antigravity
# Installs to ~/.gemini/antigravity/skills/ by default
```
=== "Manual"
```bash
cp -R integrations/antigravity/* ~/.gemini/antigravity/skills/
```
Verify
find ~/.gemini/antigravity/skills -name "SKILL.md" | wc -l
# Expected: 156
Hermes Agent
Hermes Agent by Nous Research is a self-improving AI agent with a built-in learning loop. It uses the agentskills.io standard — the same SKILL.md format our repo uses — so no conversion is needed.
!!! tip "Install-tested in CI"
Every pull request that touches skills installs the real hermes-agent package into an empty home directory, runs scripts/sync-hermes-skills.py, and fails unless hermes skills list shows every installed skill ({{ count.hermes }} today). See harness-compat.yml.
Why Hermes is different
Unlike other tools that need format conversion, Hermes reads SKILL.md files natively with the exact same YAML frontmatter (name, description, version, license), the same directory layout (references/, templates/, assets/), and the same AGENTS.md project context. Our skills are plug-and-play.
Step 1 — Install Hermes Agent itself
python3 -m venv ~/.venvs/hermes
~/.venvs/hermes/bin/pip install hermes-agent
export PATH="$HOME/.venvs/hermes/bin:$PATH"
hermes --version
Hermes supports several model providers (Nous, OpenAI, Anthropic and OpenAI-compatible endpoints). Configure yours as described in the Hermes Agent README.
Step 2 — Install our skills into Hermes
=== "Sync script (recommended)"
```bash
git clone https://github.com/alirezarezvani/claude-skills.git
cd claude-skills
python scripts/sync-hermes-skills.py --verbose
```
This symlinks {{ count.hermes }} skills into `~/.hermes/skills/claude-skills/<domain>/<skill>/`, where Hermes discovers them automatically. Re-run it after `git pull` to pick up new skills.
=== "Single domain"
```bash
python scripts/sync-hermes-skills.py --domain engineering --verbose
```
=== "Copy mode (portable)"
```bash
python scripts/sync-hermes-skills.py --copy --verbose
```
Creates full copies instead of symlinks. Use this on systems where symlinks across filesystems don't work, or to share with Docker containers.
=== "Manual (any single skill)"
```bash
# Symlink
ln -s /path/to/claude-skills/engineering/karpathy-coder ~/.hermes/skills/karpathy-coder
# Or copy
cp -r /path/to/claude-skills/engineering/llm-wiki ~/.hermes/skills/llm-wiki
```
Using skills in Hermes
Once installed, skills are available through Hermes's standard discovery:
/skills # Browse all installed skills (ours show up under claude-skills/)
/<skill-name> # Invoke any skill directly as a slash command
/skills search karpathy # Search by keyword
Hermes's skill_view tool loads the SKILL.md content into the conversation context, just like Claude Code does. Python scripts in scripts/ subdirectories run natively since Hermes has a full Python runtime.
What works
| Feature | Status | Notes |
|---|---|---|
| SKILL.md loading | ✅ | Identical frontmatter format (agentskills.io) |
Python scripts (scripts/) |
✅ | All stdlib-only, Hermes has Python runtime |
| References / templates / assets | ✅ | Same directory convention |
AGENTS.md project context |
✅ | Hermes reads AGENTS.md natively |
Slash commands (/<name>) |
✅ | Auto-discovered from SKILL.md |
| Sub-agents | ⚠️ | Hermes uses its own delegate_tool, not Claude Code's Agent tool — agent .md files load as context but dispatch mechanism differs |
| Claude Code plugin.json | ➖ | Hermes ignores this — not needed, it scans SKILL.md directly |
| Hooks (settings.json) | ⚠️ | Different hook system — manual wiring for Hermes's config.yaml |
Verify
hermes skills list | grep -c claude-skills
# Expected: {{ count.hermes }}
Updating
cd claude-skills
git pull origin main
python scripts/sync-hermes-skills.py --verbose
# Existing symlinks are preserved, new skills are added
Step 3 — First-run walkthrough
A complete dry-run from cold install to running your first skill:
# 1. After Steps 1 + 2 above (Hermes installed, skills synced)
hermes
# 2. Check what skills are loaded
> /skills
# → claude-skills/engineering/karpathy-coder
# → claude-skills/research/research
# → claude-skills/productivity/capture
# ... ({{ count.hermes }} total)
# 3. Invoke a skill — research orchestrator example
> /research What's the state of post-quantum cryptography in 2026?
# → Hermes loads research/research/SKILL.md, runs Q1+Q2 intake,
# classifies via SIGNALS map, routes to research-pack specialist
# (litreview here) or runs fallback workflow
# 4. Or browse a skill's docs without running it
> /skill_view claude-skills/engineering/karpathy-coder
# → loads SKILL.md content as context; you can ask questions about it
# 5. Search for a topic
> /skills search "test generation"
# → ranks claude-skills entries by description match
Configuration tips
Edit ~/.hermes/config.yaml (created on first run) to customize how Hermes uses our skills:
# Recommended config for working with the claude-skills tree
skills:
search_paths:
- ~/.hermes/skills/ # Default location (our sync script lands here)
auto_load:
- claude-skills/engineering/karpathy-coder # Always-loaded skills (high-impact, low-token)
- claude-skills/engineering/grill-me # Use sparingly — adds intake friction
display:
show_category: true # Group results by claude-skills/<domain>/
show_description: true
Troubleshooting
??? question "/skills shows 0 results after running the sync script"
The sync target may be wrong. Check what the script actually did:
bash python scripts/sync-hermes-skills.py --target ~/.hermes/skills --verbose --dry-run # Should show {{ count.hermes }} skills queued for sync. If 0, your DOMAIN_DIRS list is wrong (regression — file an issue). ls -la ~/.hermes/skills/claude-skills/ # If empty, the sync didn't actually write — check for permission errors above.
??? question "Symlinks point to a path on someone else's machine"
You probably cloned a fork that committed absolute-path symlinks. Re-run the sync from your own clone — v2.7.2+ generates relative symlinks (../../../../<domain>/...) which work across machines:
bash rm -rf ~/.hermes/skills/claude-skills/ python scripts/sync-hermes-skills.py --verbose
??? question "Slash commands like /research collide with Hermes's built-ins"
Hermes resolves user-defined skills first, so /research from claude-skills wins. If you want the built-in instead, use the fully qualified path: /skill_view hermes/research. To avoid collisions entirely, rename via symlink: ln -s ~/.hermes/skills/claude-skills/research/research ~/.hermes/skills/cs-research.
??? question "Python tools fail with ModuleNotFoundError"
Our scripts are stdlib-only by policy — ModuleNotFoundError means either (a) you're running an old Python (we require 3.10+) or (b) the script itself violated the policy (file a bug). Confirm:
bash python3 --version # Must be ≥ 3.10 python3 ~/.hermes/skills/claude-skills/engineering/karpathy-coder/scripts/karpathy_lint.py --help # Should print help text without errors
??? question "Hermes can't find SKILL.md but the file exists"
Hermes expects SKILL.md at the top of the skill directory. Our nested-plugin layout (<domain>/<plugin>/skills/<skill>/SKILL.md) is flattened by the sync script — the symlink at ~/.hermes/skills/claude-skills/<domain>/<skill>/ points directly at the inner skills/<skill>/ folder, so SKILL.md is at the top level after the symlink jump. If a specific skill is missing, check:
bash ls -la ~/.hermes/skills/claude-skills/<domain>/<skill>/SKILL.md # If "No such file", the symlink target is broken — re-run the sync script
??? question "How do I unsync (remove our skills from Hermes)?"
bash rm -rf ~/.hermes/skills/claude-skills/ # Hermes's own built-in skills are unaffected (they live elsewhere in ~/.hermes/skills/)
Mistral Vibe
Mistral Vibe is Mistral AI's open-source Apache-2.0 CLI coding agent (v2.0, released January 2026). It uses the Agent Skills standard — the same SKILL.md + YAML frontmatter format Claude Code and Hermes Agent use — so no conversion is needed.
!!! tip "Install-tested in CI"
Every pull request that touches skills installs the real mistral-vibe package into an empty home directory, runs scripts/vibe-install.sh, and fails unless Vibe's own skill manager loads every installed skill ({{ count.vibe }} today). See harness-compat.yml.
Discovery paths
Per the official docs, Vibe scans three locations for skills:
| Path | Scope |
|---|---|
~/.vibe/skills/ |
User-global (what our sync script writes to) |
.vibe/skills/ |
Project-local |
.agents/skills/ |
Agent Skills standard path |
Step 1 — Install Mistral Vibe itself
If you don't have Vibe installed yet, follow the Vibe quickstart:
pip install mistral-vibe
vibe --version # Verify install
Vibe supports both Mistral's hosted models (via MISTRAL_API_KEY) and self-hosted endpoints. See the Vibe CLI docs for provider configuration.
Step 2 — Install our skills into Vibe
=== "Sync script (recommended)"
```bash
git clone https://github.com/alirezarezvani/claude-skills.git
cd claude-skills
./scripts/vibe-install.sh
```
This symlinks {{ count.vibe }} skills into `~/.vibe/skills/<skill>/`, the flat user-global path Vibe scans by default. (The old `--nested` layout under `claude-skills/<domain>/` is only discovered if you add `skill_paths` to `~/.vibe/config.toml`.)
=== "Single domain"
```bash
python scripts/sync-vibe-skills.py --domain engineering --verbose
```
=== "Copy mode (portable)"
```bash
python scripts/sync-vibe-skills.py --copy --verbose
```
Creates full copies instead of symlinks — useful for Docker containers or shared filesystems where symlinks don't traverse cleanly.
=== "Custom target"
```bash
python scripts/sync-vibe-skills.py --target /opt/team-vibe/skills/
```
Useful for team-wide installs or sandboxed environments.
Using skills in Vibe
Once installed, skills are available through Vibe's standard discovery (per the Vibe Agents & Skills docs):
/skills # List all installed skills
/<skill-name> # Invoke a skill by slug as a slash command
Vibe can also auto-load skills when your prompt matches a skill's description field — same trigger mechanism Claude Code uses.
What works
| Feature | Status | Notes |
|---|---|---|
| SKILL.md loading | ✅ | Identical YAML frontmatter (agentskills.io) |
Python scripts (scripts/) |
✅ | All stdlib-only; Vibe shell tool runs them |
| References / templates / assets | ✅ | Same directory convention |
Slash commands (/<name>) |
✅ | Auto-discovered from SKILL.md |
| Sub-agents | ⚠️ | Vibe uses its own subagent system with TOML configs (~/.vibe/agents/) — Claude Code agent .md files load as context but dispatch differs |
| Claude Code plugin.json | ➖ | Vibe ignores this — scans SKILL.md directly |
| Hooks (settings.json) | ➖ | Vibe has its own hook system; manual wiring required |
Verify
ls ~/.vibe/skills | wc -l # one folder per skill
# Expected: about {{ count.vibe }}
# Or in the Vibe CLI
vibe
> /skills
Updating
cd claude-skills
git pull origin main
python scripts/sync-vibe-skills.py --verbose
# Existing symlinks are preserved, new skills are added
Troubleshooting
??? question "Vibe doesn't see the synced skills"
Make sure the sync ran and the symlinks resolve:
bash ls -la ~/.vibe/skills/agent-designer/SKILL.md # Should print a valid SKILL.md, not "No such file"
If symlinks are broken (the source repo was moved), re-run python scripts/sync-vibe-skills.py --verbose.
??? question "How do I unsync (remove our skills from Vibe)?"
bash python3 scripts/sync-vibe-skills.py --dry-run --json # lists the skill folders it manages # then remove those folders from ~/.vibe/skills/
Script Reference
convert.sh
Usage:
./scripts/convert.sh [--tool <name>] [--out <dir>] [--help]
Tools:
antigravity, cursor, aider, kilocode, windsurf, opencode, augment, all
Options:
--tool <name> Convert for a specific tool (default: all)
--out <dir> Output directory (default: integrations/)
--help Show usage
Examples:
# Convert all skills for all tools
./scripts/convert.sh
# Convert only for Cursor
./scripts/convert.sh --tool cursor
# Custom output directory
./scripts/convert.sh --tool windsurf --out /tmp/my-skills
install.sh
Usage:
./scripts/install.sh --tool <name> [--target <dir>] [--force] [--help]
Options:
--tool <name> Required. Which tool to install for.
--target <dir> Project directory (default: current dir, except antigravity)
--force Skip overwrite confirmation
--help Show usage
Default install locations:
| Tool | Default Target |
|---|---|
| Hermes Agent | ~/.hermes/skills/claude-skills/ |
| Antigravity | ~/.gemini/antigravity/skills/ |
| Cursor | <target>/.cursor/rules/ |
| Aider | <target>/CONVENTIONS.md |
| Kilo Code | <target>/.kilocode/rules/ |
| Windsurf | <target>/.windsurf/skills/ |
| OpenCode | <target>/.opencode/skills/ |
| Augment | <target>/.augment/rules/ |
Troubleshooting
??? question "I get 'No skills found' when running convert.sh" Make sure you're running the script from the repository root where the skill directories are located.
??? question "Some skills show garbled descriptions"
This can happen with skills using complex YAML multi-line descriptions. Re-run convert.sh — the parser handles folded (>) and literal (|) YAML scalars.
??? question "Can I use skills from multiple tools at once?" Yes! You can install skills for Cursor and Windsurf in the same project — they use different directories and won't conflict.
??? question "How do I update when new skills are added?"
bash git pull origin main ./scripts/convert.sh --tool all ./scripts/install.sh --tool <your-tool> --target . --force
??? question "Can I convert only specific skills?"
Not yet via CLI flags, but you can run convert.sh and then copy only the skills you want from integrations/<tool>/.
??? question "Do supporting files (scripts, references) work in all tools?" Only tools that support subdirectories per skill (Hermes Agent, Antigravity, Windsurf, OpenCode) get the full bundle. Flat-file tools (Cursor, Aider, Kilo Code, Augment) get the SKILL.md content only.
??? question "Does Hermes Agent need format conversion?"
No. Hermes uses the same agentskills.io SKILL.md format as our repo. Just run python scripts/sync-hermes-skills.py --verbose to symlink skills into ~/.hermes/skills/. No conversion step needed.