#Skill Guide

OmniLLM ships with a first-party agent skill in the repository's skill/ directory. The skill teaches agents the crate's real boundaries:

• runtime generation through Gateway, ProviderEndpoint, and EndpointProtocol
• provider primitive runtime calls through PrimitiveRequest, PrimitiveProviderEndpoint, and Gateway::primitive_*
• protocol parsing, emission, and transcoding through parse_*, emit_*, and transcode_*
• typed multi-endpoint conversion through ApiRequest, ApiResponse, and WireFormat
• replay fixture sanitization through ReplayFixture and sanitize_*

If you only need the Rust crate, go back to Usage Guide. This page is specifically about installing the OmniLLM Skill into coding agents.

#Install With Vercel Labs Skills

These instructions use the Vercel Labs skills installer.

The skill is declared as omnillm. When you install with --skill omnillm, the installer creates the correct target directory name automatically.

Agent runtimes only require:

• SKILL.md
• references/
• assets/

The installer may also add README.md next to the skill files and a project-level skills-lock.json.

The commands below install directly from GitHub, so you do not need to clone the repository first.

The commands below use --copy so the installed skill stays self-contained in the target agent directory.

#Claude Code

npx skills add https://github.com/aiomni/omnillm --skill omnillm --agent claude-code --copy

Add -g for a user-level install.

#Codex

npx skills add https://github.com/aiomni/omnillm --skill omnillm --agent codex --copy

Add -g for a user-level install.

#OpenCode

npx skills add https://github.com/aiomni/omnillm --skill omnillm --agent opencode --copy

Add -g for a user-level install.

#Verify Installation

Use the installer to confirm that the skill is present for the agent you care about:

npx skills ls -a codex --json

Replace codex with claude-code or opencode as needed.

Then start a new session in your chosen agent and ask for something OmniLLM-specific, for example:

• scaffold a GatewayBuilder flow with ProviderEndpoint and KeyConfig
• configure an EndpointProtocol::*_compat runtime endpoint for an OpenAI-compatible wrapper that requires messages[].content[]
• debug an OpenAI Chat compat stream where delta.role and the first delta.content arrive in the same SSE frame
• pass through wrapper-specific OpenAI top-level fields such as enable_thinking with LlmRequest.vendor_extensions
• explain when canonical Gateway APIs, provider primitive APIs, or transcode_* are correct
• route a provider-native PrimitiveRequest through primitive_call, primitive_stream, or primitive_realtime
• debug NoAvailableKey, BudgetExceeded, or Protocol(...)
• emit an ApiRequest into a provider wire format

If the skill does not appear immediately, restart the session and rerun npx skills ls -a <agent>.