Targets

View as MarkdownThe same page as plain textOpen in ChatGPTAsk about this pageOpen in ClaudeAsk about this page

The 77 runtimes, the two scopes, and where each artifact lands on disk.

Where a skill lands is a function of three things, the artifact type, the runtime you named, and the scope you are writing in. This page covers the last two.

The two scopes

Project is the default, and it only applies inside a git repository. Outside one, a plain flag line refuses, exit 2, because the rule this product follows is that the git is the manifest: nothing else on a machine can audit what a silent write left behind. Global, with --global or -g, writes under the home directory instead, and needs no git at all.

The two scopes are not the same operation at a different root. They write differently.

	Project	Global
What lands	a real copy, always	a real copy at the first destination, relative links after it
Needs git	yes, and refuses outside one	no
Safety net	git status reveals and undoes	none, which is why --force exists
Occupied destination	overwritten, because git shows it	asks in a terminal; refused with exit 3 off a terminal, under --yes or under --dry-run; --force overwrites without asking

The two scopes, seen from disk

The table above gives the rule; this is what it produces. One skill installed for two runtimes that share the same folder:

# project scope, inside the repository
.claude/skills/panlabs-python-standards/
├── SKILL.md
└── references/...

# machine scope, under the home directory
~/.claude/skills/panlabs-python-standards/     ← real copy, the first destination
~/.agents/skills/panlabs-python-standards  ->  ../../.claude/skills/panlabs-python-standards

The second destination onward is a relative link, and relative on purpose: a home directory that moves, or that is synced across machines, keeps resolving.

doctor knows the difference and reads it from both sides. A link pointing at nothing is a dangling link; a link that turned into a text file, because someone copied the folder without preserving links, is a link turned into text. Both are findings, and a finding fails.

In project scope, every landing is a real copy, never a symlink. This is deliberate: under core.symlinks=false, a git config value a clone can inherit silently, a committed symlink checks out as an ordinary text file instead of a link, and the equipment would be broken for anyone who cloned it. Copying sidesteps that failure mode entirely.

In global scope, each selection climbs a small ladder: the first destination in the runtime table's own order receives a real copy, and every destination after it becomes a relative symlink pointing at that first copy. Relative, specifically so the link keeps working if $HOME itself moves. On Windows the same role is filled by a junction, which needs no elevated privilege. Where neither can be created, the write silently degrades to a second real copy, prints a warning saying so, and still exits 0.

The universal group

Several runtimes read the exact same directory. In project scope, 19 of the 77, including the synthetic universal key itself, read .agents/skills, which makes it the single largest shared destination. Despite the name it is not a path that reaches every runtime: measured against Claude Code 2.1.223, it does not discover skills there, and Codex does not read .claude/skills either.

In global scope the same group shrinks to 6. cline, dexto, kimi-code-cli, loaf, warp and zed all read ~/.agents/skills, while amp reads ~/.config/agents/skills on its own XDG-derived path, codex reads ~/.codex/skills, and cursor reads ~/.cursor/skills.

The plan always names every destination and who reads it, so a selection landing in a shared folder is shown as one write serving several runtimes, never as a promise of one install per runtime.

The runtime table

77 rows: 76 transcribed from the upstream vercel-labs/skills project's own runtime map, attributed in NOTICE, plus vscode, which has no skill destination in either scope and exists in this table purely so the MCP graft step of the wizard has a name for it.

74 of the 77 have a skills destination in global scope. eve and promptscript do not, so asking for either under --global is a valid line with a negative answer, exit 3, rather than an invalid one.

vscode is the third, and it refuses on a different axis: it has no skills destination in either scope, and the refusal does not change with --global. The message is different too, and names what it takes instead:

`vscode` takes MCP servers and has no skills destination of its own;

--runtime vscode --mcp <name> installs; --runtime vscode --skill <name> exits 3 in both scopes.

Global paths below use ~ for the home directory. Several are actually resolved through an environment variable first, falling back to the path shown only when that variable is unset. There are three mechanisms, and the table does not tell them apart:

• six honour a variable of the tool's own: autohand-code (AUTOHAND_HOME), claude-code (CLAUDE_CONFIG_DIR), codex (CODEX_HOME), grok (GROK_HOME), hermes-agent (HERMES_HOME) and mistral-vibe (VIBE_HOME);
• six resolve through XDG_CONFIG_HOME before falling back to ~/.config, and they are amp, devin, goose, opencode, replit and universal;
• one uses the first directory that exists: openclaw looks for ~/.openclaw, ~/.clawdbot and ~/.moltbot in that order, and falls back to the first when none exists.

The other ~/.config paths in the table are literals, not XDG: crush and kimchi live there by direct spelling, and do not move with the variable.

A blank value counts as unset. A relative value is ignored on purpose, and the tool declares that divergence against upstream.

--runtime key	Runtime	Project scope	Global scope
aider-desk	AiderDesk	.aider-desk/skills	~/.aider-desk/skills
amp	Amp	.agents/skills	~/.config/agents/skills
antigravity	Antigravity	.agents/skills	~/.gemini/antigravity/skills
antigravity-cli	Antigravity CLI	.agents/skills	~/.gemini/antigravity-cli/skills
astrbot	AstrBot	data/skills	~/.astrbot/data/skills
autohand-code	Autohand Code CLI	.autohand/skills	~/.autohand/skills
augment	Augment	.augment/skills	~/.augment/skills
bob	IBM Bob	.bob/skills	~/.bob/skills
claude-code	Claude Code	.claude/skills	~/.claude/skills
openclaw	OpenClaw	skills	~/.openclaw/skills
cline	Cline	.agents/skills	~/.agents/skills
codearts-agent	CodeArts Agent	.codeartsdoer/skills	~/.codeartsdoer/skills
codebuddy	CodeBuddy	.codebuddy/skills	~/.codebuddy/skills
codemaker	Codemaker	.codemaker/skills	~/.codemaker/skills
codestudio	Code Studio	.codestudio/skills	~/.codestudio/skills
codex	Codex	.agents/skills	~/.codex/skills
command-code	Command Code	.commandcode/skills	~/.commandcode/skills
continue	Continue	.continue/skills	~/.continue/skills
cortex	Cortex Code	.cortex/skills	~/.snowflake/cortex/skills
crush	Crush	.crush/skills	~/.config/crush/skills
cursor	Cursor	.agents/skills	~/.cursor/skills
deepagents	Deep Agents	.agents/skills	~/.deepagents/agent/skills
devin	Devin for Terminal	.devin/skills	~/.config/devin/skills
dexto	Dexto	.agents/skills	~/.agents/skills
droid	Droid	.factory/skills	~/.factory/skills
eve	Eve	agent/skills	none
firebender	Firebender	.agents/skills	~/.firebender/skills
forgecode	ForgeCode	.forge/skills	~/.forge/skills
gemini-cli	Gemini CLI	.agents/skills	~/.gemini/skills
github-copilot	GitHub Copilot	.agents/skills	~/.copilot/skills
goose	Goose	.goose/skills	~/.config/goose/skills
grok	Grok Build	.grok/skills	~/.grok/skills
hermes-agent	Hermes Agent	.hermes/skills	~/.hermes/skills
inference-sh	inference.sh	.inferencesh/skills	~/.inferencesh/skills
jazz	Jazz	.jazz/skills	~/.jazz/skills
junie	Junie	.junie/skills	~/.junie/skills
iflow-cli	iFlow CLI	.iflow/skills	~/.iflow/skills
kilo	Kilo Code	.kilocode/skills	~/.kilocode/skills
kimchi	Kimchi	.kimchi/skills	~/.config/kimchi/harness/skills
kimi-code-cli	Kimi Code CLI	.agents/skills	~/.agents/skills
kiro-cli	Kiro CLI	.kiro/skills	~/.kiro/skills
kode	Kode	.kode/skills	~/.kode/skills
lingma	Lingma	.lingma/skills	~/.lingma/skills
loaf	Loaf	.agents/skills	~/.agents/skills
mcpjam	MCPJam	.mcpjam/skills	~/.mcpjam/skills
minimax-code	MiniMax Code	.minimax/skills	~/.minimax/skills
mistral-vibe	Mistral Vibe	.vibe/skills	~/.vibe/skills
moxby	Moxby	.moxby/skills	~/.moxby/skills
mux	Mux	.mux/skills	~/.mux/skills
opencode	OpenCode	.agents/skills	~/.config/opencode/skills
openhands	OpenHands	.openhands/skills	~/.openhands/skills
ona	Ona	.ona/skills	~/.ona/skills
pi	Pi	.pi/skills	~/.pi/agent/skills
qoder	Qoder	.qoder/skills	~/.qoder/skills
qoder-cn	Qoder CN	.qoder/skills	~/.qoder-cn/skills
qwen-code	Qwen Code	.qwen/skills	~/.qwen/skills
replit	Replit	.agents/skills	~/.config/agents/skills
reasonix	Reasonix	.reasonix/skills	~/.reasonix/skills
rovodev	Rovo Dev	.rovodev/skills	~/.rovodev/skills
roo	Roo Code	.roo/skills	~/.roo/skills
tabnine-cli	Tabnine CLI	.tabnine/agent/skills	~/.tabnine/agent/skills
terramind	Terramind	.terramind/skills	~/.terramind/skills
tinycloud	Tinycloud	.tinycloud/skills	~/.tinycloud/skills
trae	Trae	.trae/skills	~/.trae/skills
trae-cn	Trae CN	.trae/skills	~/.trae-cn/skills
warp	Warp	.agents/skills	~/.agents/skills
windsurf	Windsurf	.windsurf/skills	~/.codeium/windsurf/skills
zed	Zed	.agents/skills	~/.agents/skills
zcode	ZCode	.zcode/skills	~/.zcode/skills
zencoder	Zencoder	.zencoder/skills	~/.zencoder/skills
zenflow	Zenflow	.zencoder/skills	~/.zencoder/skills
neovate	Neovate	.neovate/skills	~/.neovate/skills
pochi	Pochi	.pochi/skills	~/.pochi/skills
promptscript	PromptScript	.agents/skills	none
adal	AdaL	.adal/skills	~/.adal/skills
vscode	VS Code	none	none
universal	Universal	.agents/skills	~/.config/agents/skills

vscode has no row in either column because it is not a skill destination at all. It exists in the table only for the MCP graft, which reads and writes .vscode/mcp.json and nothing under skills/. universal is not a real tool: it is a synthetic key naming the shared .agents/skills destination directly, for when you want to equip that folder without naming every runtime that happens to read it.