Target Surface Support Matrix
Support claims are intentionally separate from skill selection. This generated page separates install eligibility from support claims. supported_agents in manifest/skills.yaml and manifest/artifacts.yaml is selection eligibility; the rows below state how each target surface is rendered, whether it is supported, degraded, blocked, manual, or unsupported, and what code or policy backs that claim. Do not infer runtime support from supported_agents alone.
The current manifest contains 59 installable skills; the matrix below describes the target/surface support contract that those generated skill and artifact plans use.
OpenClaw source/import evidence and OpenClaw install-target behavior are separate. Source/import evidence never authorizes real .openclaw writes; current OpenClaw install-target behavior remains fake-root scoped.
Target |
Surface |
State |
Mechanism |
Scope |
Claim Basis |
Notes |
|---|---|---|---|---|---|---|
|
|
|
|
agent-visible regular SKILL.md file with copied support files |
|
Auto mode copies canonical skills and support files because symlinked Codex skill loading is not assumed and installed skills must remain self-contained. |
|
|
|
|
native Claude skill file |
|
Auto mode links to canonical SKILL.md with apply-time fallback. |
|
|
|
|
agent-visible regular SKILL.md adapter pointing at canonical source |
|
Auto mode uses reference adapters because native symlink loading is not verified. |
|
|
|
|
personal GitHub Copilot skill adapter |
|
Copilot receives personal skill/persona surfaces only; commands/templates are unsupported. |
|
|
|
|
OpenCode-native regular SKILL.md file with copied support files |
|
Auto mode copies canonical skills for cross-platform parity; explicit reference and symlink modes remain available with evidence. |
|
|
|
|
Antigravity global flat Markdown skill file with the embedded canonical body |
|
Auto mode writes |
|
|
|
|
fake-root install-target layout only |
|
OpenClaw real-system target writes stay blocked until native target evidence exists. |
|
|
|
|
non-executing quick-action reference |
|
Codex does not receive native slash-command registrations from this artifact class. |
|
|
|
|
Claude command file |
|
Claude is the only current target that receives native command files for entrypoint aliases. |
|
|
|
|
non-executing quick-action reference |
|
DeepSeek receives entrypoint references rather than native command registrations. |
|
|
|
|
not installed |
|
Copilot optional artifacts are limited to supported personal surfaces. |
|
|
|
|
OpenCode command markdown file |
|
OpenCode receives native command aliases under ~/.config/opencode/commands. |
|
|
|
|
Antigravity global flat Markdown skill alias |
|
Entry-point aliases render as global Antigravity Markdown skills under ~/.gemini/antigravity-cli/skills/ (or ~/.gemini/config/skills/ on a home the vendor has migrated). |
|
|
|
|
not installed |
|
OpenClaw commands/aliases remain outside target support. |
|
|
|
|
root-scoped managed runtime helpers |
|
Runtime files are copied under the selected root runtime, not per-agent skill folders. |
|
|
|
|
root-scoped managed runtime helpers |
|
Runtime files are copied under the selected root runtime, not per-agent skill folders. |
|
|
|
|
root-scoped managed runtime helpers |
|
Runtime files are copied under the selected root runtime, not per-agent skill folders. |
|
|
|
|
root-scoped managed runtime helpers |
|
Runtime files are copied under the selected root runtime, not per-agent skill folders. |
|
|
|
|
root-scoped managed runtime helpers |
|
Runtime files are copied under the neutral shared ai-agents-skills runtime root, not inside OpenCode config. |
|
|
|
|
root-scoped managed runtime helpers |
|
Runtime files use the neutral shared ai-agents-skills runtime root, not a Codex-specific runtime path. |
|
|
|
|
fake-root by default; real-system gated by an approved manifest + host broker |
|
Real-system runtime files install via an approved content-addressed runtime manifest (openclaw-runtime-apply-manifest –real-system + confirmation phrase, verify-before-write): inert files under .openclaw/skills/ |
|
|
|
|
Antigravity plugin agent definition |
|
Personas are installed under ~/.gemini/antigravity-cli/plugins/ai-agents-skills/ (or ~/.gemini/config/plugins/ai-agents-skills/ on a home the vendor has migrated) in an agents/ subdirectory, with the plugin manifest. |
|
|
|
|
Antigravity native plugin package |
|
The installer creates plugin.json and a managed plugin payload under ~/.gemini/antigravity-cli/plugins/ai-agents-skills/ (or ~/.gemini/config/plugins/ai-agents-skills/ on a home the vendor has migrated). |
|
|
|
|
managed block in ~/.gemini/GEMINI.md |
|
Skill and repo-management instruction blocks use the documented Antigravity global developer context file. |
|
|
|
|
sparse Antigravity settings JSON |
|
A no-op settings.json scaffold is managed under ~/.gemini/antigravity-cli/settings.json when Antigravity artifacts are installed. |
|
|
|
|
plugin-scoped Antigravity MCP config |
|
A no-op mcp_config.json scaffold with an empty mcpServers map is installed inside the managed Antigravity plugin. |
|
|
|
|
plugin-scoped Antigravity hooks config |
|
A no-op hooks.json scaffold is installed inside the managed Antigravity plugin. |
|
|
|
|
managed Stop-hook entry merged into ~/.claude/settings.json |
|
When the autonomous-research-loop runtime is installed, one managed hooks.Stop entry (tagged _managedBy/_id) is idempotently merged into the user’s settings.json and removed on uninstall; user-authored hooks are preserved. |
|
|
|
|
Grok-native directory-layout SKILL.md file with copied support files |
|
Auto mode writes ~/.grok/skills/ |
|
|
|
|
Grok command markdown file |
|
Entry-point aliases render as native Grok command files under ~/.grok/commands/. |
|
|
|
|
root-scoped managed runtime helpers |
|
Runtime files use the neutral shared ai-agents-skills runtime root, not a Codex-specific runtime path. |
|
|
|
|
Grok subagent Markdown definition |
|
Personas install under ~/.grok/agents/ as name/description overlays; Claude-format tool-restriction frontmatter is not enforced on Grok. |
|
|
|
|
managed block in ~/.grok/AGENTS.md |
|
Skill and repo-management instruction blocks use the documented Grok home-scope AGENTS.md context file; there is no GROK.md. |
|
|
|
|
Grok rules-directory Markdown file |
|
Instruction docs copy to ~/.grok/rules/; home-scope rules/ loading is unverified. |
|
|
|
|
inert managed template storage |
|
Templates copy to ~/.grok/templates/ as inert support storage referenced by skill relative paths; not a Grok-loaded surface. |
|
|
|
|
inert managed tool storage |
|
Tool shims copy to ~/.grok/tools/ as inert support storage referenced by skill relative paths; not a Grok-loaded surface. |
|
|
|
|
discrete managed ~/.grok/hooks/ai-agents-skills-autoloop.json |
|
The optional autoloop Stop hook installs as a fully-owned native hook file under ~/.grok/hooks/ (10-hooks.md); ~/.grok/settings.json is never written. |
|
|
|
|
managed [compat.claude] block merged into ~/.grok/config.toml |
|
A managed [compat.claude] block (skills/agents/rules/hooks = false) is idempotently merged into ~/.grok/config.toml so Grok presents a single self-contained view; the block is removed on uninstall and user-authored TOML is preserved. |
|
|
|
|
Kimi Code directory-layout SKILL.md with copied support files |
|
Auto mode writes ~/.kimi-code/skills/ |
|
|
|
|
not installed |
|
Kimi has no commands/ loader; skills are invoked as /skill: |
|
|
|
|
root-scoped managed runtime helpers |
|
Runtime files use the neutral shared ai-agents-skills runtime root, not a Kimi-specific runtime path. |
|
|
|
|
Kimi custom agent Markdown definition |
|
Personas install under ~/.kimi-code/agents/ as name/description Markdown; unknown frontmatter fields are ignored by Kimi. |
|
|
|
|
managed block in ~/.kimi-code/AGENTS.md |
|
Skill and repo-management instruction blocks use the documented Kimi home-scope AGENTS.md file. |
|
|
|
|
inert managed instruction storage |
|
Instruction docs copy to ~/.kimi-code/instructions/ as inert support storage; not a documented Kimi-loaded surface. |
|
|
|
|
inert managed template storage |
|
Templates copy to ~/.kimi-code/templates/ as inert support storage referenced by skill relative paths. |
|
|
|
|
inert managed tool storage |
|
Tool shims copy to ~/.kimi-code/tools/ as inert support storage referenced by skill relative paths. |
|
|
|
|
host directory-layout SKILL.md with copied support files |
|
Auto mode writes ~/.chatgpt-local-coder/skills/ |
|
|
|
|
non-executing quick-action reference |
|
The host exposes skills through skill_list/skill_run and has no slash-command loader, so aliases install as reference docs under ~/.chatgpt-local-coder/instructions/entrypoints/ instead of as native commands. |
|
|
|
|
root-scoped managed runtime helpers |
|
Runtime files use the neutral shared ai-agents-skills runtime root, not a host-specific runtime path. |
|
|
|
|
inert managed persona storage |
|
Personas copy to ~/.chatgpt-local-coder/agents/ as name/description Markdown. The host imports subagent definitions from ~/.claude/agents only; it does not yet read its own agents directory, so these files are reference material rather than registered subagents. |
|
|
|
|
managed block in ~/.chatgpt-local-coder/AGENTS.md |
|
The host loads ~/.chatgpt-local-coder/AGENTS.md as its user-level memory file, so managed skill and repo-management blocks reach the model. |
|
|
|
|
inert managed instruction storage |
|
Instruction docs copy to ~/.chatgpt-local-coder/instructions/ as inert support storage referenced by skill relative paths; the host does not auto-load that directory. |
|
|
|
|
inert managed template storage |
|
Templates copy to ~/.chatgpt-local-coder/templates/ as inert support storage referenced by skill relative paths. |
|
|
|
|
inert managed tool storage |
|
Tool shims copy to ~/.chatgpt-local-coder/tools/ as inert support storage referenced by skill relative paths; they are not registered as MCP tools. |
Claim levels used here:
supported: installer behavior is implemented for the listed surface.fallbackordegraded: the installer can proceed with reduced or apply-time fallback behavior.blocked: the installer intentionally refuses the surface.manual: the surface needs explicit fake-root/manual evidence and is not a real-system support claim.unsupported: the target does not receive that surface today.