Add Wallaby provider (wallaby/kimi-k3)

[Dawncoral]

Adds Wallaby as an OpenAI-compatible provider serving moonshotai/kimi-k3.

• Endpoint: https://api.wallabytoken.com/v1
• Docs: https://www.wallabytoken.com/docs
• Cost: $2.70 input / $13.50 output / $0.27 cached read per 1M tokens (current pricing)

Only overrides cost; all model facts come from base_model moonshotai/kimi-k3.

[github-actions]

Action items

• [high] [violation] providers/wallaby/logo.svg:1 - Check: New provider logo.svg must use currentColor, have no fixed width/height, and preferably a square viewBox. Why: The SVG sets width="24" height="24", which violates the no-fixed-size logo rule and is a merge blocker for new providers. Action: Remove the fixed width and height attributes; keep a square viewBox (e.g. 0 0 24 24) and currentColor stroke/fill only.
• [high] [violation] providers/wallaby/models/moonshotai/kimi-k3.toml:1 - Check: Provider models with reasoning = true (inherited from base_model) must set host reasoning_options per lab/peer baseline. Why: moonshotai/kimi-k3 is a reasoner; Wallaby is a third-party OpenAI-compatible host and only authors cost. Lab and same-surface peers (e.g. Moonshot, OpenRouter, TokenGo) expose caller controls (toggle + effort low/high/max, or equivalent). Omitting reasoning_options is invalid and understates API controls. Action: Add reasoning_options matching this host’s wire surface and the Kimi K3 lab/peer baseline (typically toggle + effort ["low", "high", "max"]), with a leading top-of-file comment documenting the exact toggle/effort request fields if a toggle is used.
• [low] [possible mistake] providers/wallaby/models/moonshotai/kimi-k3.toml:3 - Check: Data-changing PRs should cite direct provider pricing/docs mapped to claims. Why: Cost is set to $2.70 / $13.50 / $0.27 cache_read with only a generic docs URL in the PR body, so the rates cannot be reviewed against a first-party pricing source. Action: Cite the Wallaby pricing page (or equivalent) and note what each URL supports for input/output/cache_read.

[github-actions]

Action items

• [high] [violation] providers/wallaby/logo.svg:1 - Check: New provider logo.svg must use currentColor, have no fixed size, and no hardcoded colors (AGENTS.md Logo). Why: The root <svg> sets width="24" height="24", which is a fixed size and fails the new-provider logo blocker. Action: Drop the width and height attributes; keep a square viewBox (e.g. 0 0 24 24) and stroke="currentColor" only.
• [high] [possible mistake] providers/wallaby/models/moonshotai/kimi-k3.toml:7 - Check: Relay reasoning_options must follow the lab + same-surface peer baseline for the underlying model, not invent a GPT-style set (AGENTS.md Reasoning options; audit skill). Why: First-party providers/moonshotai/models/kimi-k3.toml and solid peers (OpenRouter, kimi-for-coding, Novita, etc.) use effort low / high / max (often with a toggle). This host declares low / medium / high, adding medium (which peers such as aiand report K3 rejects) and dropping lab max. A comment asserting host verification is not enough against that contradiction. Action: Re-probe this host and align options with what actually works: prefer lab/peer ["low", "high", "max"] (plus toggle only if a real on/off wire path exists), or keep a divergent set only with concrete probe evidence that medium is a distinct graded level and that max is unsupported—not merely accepted.

[github-actions]

No actionable findings.