How long does the integration take?

Change one base URL in an OpenAI-compatible client, or install the Python or TypeScript SDK; the quickstart routes a first request in a few minutes.

Will this lock us into another vendor?

No. The router is Apache-2.0 open source, your provider keys stay yours, and the API is OpenAI-compatible, so leaving is a change of base URL.

What happens when a provider fails?

The router falls back to the next model your policy allows and records the fallback in the trace, so an outage costs a retry, not an incident.

pixelwright - Skill - OpenSmartRoute

Imported from m4bwav/pixelwright (AGENTS.md). Install upstream with npx skills add m4bwav/pixelwright. Copyright stays with the author.

Agent rules for pixelwright

Rules for any AI agent (Claude Code, Copilot, Cursor, Codex, Gemini CLI) working in this repository. CLAUDE.md imports this file.

Read README.md first. Three skills live in skills/: the hub pixelwright (with the one CLI, scripts/pxw.py), pixelwright-make and pixelwright-animate (with the sw.py shim for the old name). The evergreen unit sits at the repository root: RESEARCH, CHANGELOG, LEARNINGS, TESTS, evals/evals.json, MAINTENANCE and evergreen.json.

Rules

Evergreen unit in pointer mode: before editing a skill, read evergreen.json. If next_due has passed or contradiction is set, say so and refresh after the task. Every edit to a SKILL.md or a reference gets a CHANGELOG entry with its reason (R-, L-, T- ids, or "user request").
pxw.py stays Python 3.9+ with numpy and Pillow only, prints one JSON object per command (except review, which serves), exits 1 on a failed check or a missing input and 2 on a usage error (argparse's text; --version prints plain text), makes no network calls, and runs on Windows, macOS and Linux.
Every new effect, command or gate gets an offline test (tests/test_pxw.py for animation, tests/test_make.py for make) with its fixture drawn in code, and the test must fail when the behaviour breaks. A change to an existing effect or make step must keep its output byte-identical unless the CHANGELOG says why it changed (projects rely on reproducible sheets and sprites).
artwright-pixel is the judge: call aw.py pixel as a gate, never copy its checks into pxw.py.
No images in the repository except .github/images/banner.jpg. No project names, machine names, absolute paths or credentials in any file. tests/check_skills.py also flags every name listed in .private-names.txt (one per line, untracked); keep the client and people names of projects that use this plugin there.
Skills stay under 200 lines with a description of at most 1,024 characters that names the siblings that own nearby requests (comfyui-gen renders stills; artwright-pixel measures).
Before committing: python tests/test_pxw.py, python tests/test_make.py and python tests/check_skills.py.
Versions: when a packaged file changes, bump .claude-plugin/plugin.json and VERSION in pxw.py together, and add CHANGELOG release notes. Tags and releases need the owner's go-ahead.
No AI attribution anywhere: no AI co-author trailers and no AI credit lines in commits, pull requests or files.

Install pixelwright and it becomes one more candidate the router can pick - when it fits.

A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.

pixelwright

About

Agent rules for pixelwright

Rules

Install pixelwright and it becomes one more candidate the router can pick - when it fits.

What it costs to adopt - in engineering time.

One base URL

No vendor lock-in

Failover built in

Use it

Manifest

Reviews