agentsclimarketplace

Gateway setup

Skill JuLius5838/agw/skills/gateway-setup

Install and set up the Agent Gateway so Claude Code can use the developer's personal ChatGPT/Codex or GitHub Copilot subscription as a model backend. Use when the user says things like "set up the agent gateway", "install agw", "let me use my ChatGPT/Codex subscription in Claude Code", "use Codex as the backend", or "use Copilot models". Runs a bundled bootstrap that installs the pinned `agw` runtime with uv, then runs `agw setup`.From its SKILL.md

Install
npx -y skills add JuLius5838/agw --skill gateway-setup

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.1 KB, as published. Nobody here has run it

Agent Gateway — setup

Set up the local hybrid gateway. Native Claude requests retain Claude Code's saved subscription; configured ChatGPT/Codex or GitHub Copilot names route through a private per-developer LiteLLM process.

What to do

  1. Confirm prerequisites. The developer needs uv installed and the native Claude Code CLI on PATH. If uv is missing, point them to https://docs.astral.sh/uv/ and stop.

  2. Run the bundled bootstrap. It performs an ownership-aware preflight (it refuses to overwrite an unrelated agw), installs/upgrades the pinned runtime as a uv tool, and runs agw setup:

    bash "${CLAUDE_PLUGIN_ROOT}/scripts/bootstrap.sh"
    

    agw setup preserves user-edited registries. When an upgrade retires a provider entry, setup first saves a one-time permission-restricted rollback copy beside models.yaml. It enables the marked, reversible bare-claude integration by default when the shell is detectable; pass --no-shell to opt out.

  3. Enable the requested exact route. Candidates ship disabled so installation remains native-only until the developer makes a provider choice. Inspect them, then enable the requested model/provider pair:

    agw models list --all
    agw setup --provider-owner gpt-5.6-sol=chatgpt
    

    Use the exact name/provider requested by the developer; never silently pick a candidate or enable every provider.

  4. Authenticate that provider (interactive device flow — needs a terminal):

    agw auth chatgpt     # ChatGPT / Codex subscription
    agw auth copilot    # GitHub Copilot subscription
    

    The short-lived verification URL/code appears only on the terminal. Never copy it into logs or a message.

  5. Launch Claude through the gateway:

    claude                            # native Claude default, gateway available
    claude --model gpt-5.6-sol        # a specific external model
    

    command claude always bypasses the gateway and runs native Claude Code.

  6. Shell control: agw shell enable and agw shell disable explicitly manage the same marked block. command claude always bypasses it.

  7. Unified usage: setup installs the standalone /agw-usage command. The same dashboard is available in a terminal with agw usage.

Notes

  • The proxy binds to 127.0.0.1 only and needs a locally generated key.
  • Prompts and tool schemas are sent to the selected upstream provider under that provider's account and data policy. Say this plainly if the user asks.
  • If something looks wrong, run the gateway-doctor skill or agw doctor.

Keep looking

Skills are one crate of 627,801. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.