[Abbondanzo]

Follow-up to #77. Registers muse-wonder-1.0 explicitly so OpenCode gets its reasoning options, context limits, and multimodal capabilities instead of the generic Meta fallback.

Keeps the Responses endpoint, META_API_KEY credential path, Spark models, and forward-compatible fallback unchanged. Runtime-only: no docs, comments, or test changes.

The context size and practical output cap follow the public models.dev catalog. The output cap bounds reserved output TPM; it is not a separately published Meta maximum.

Test plan

• bun test: 749 pass, 0 fail on the final revision.
• Lint, format check, typecheck, build, LLP ref-check, and git diff --check pass.
• Live built CLI review against a local pricing fixture with meta/muse-wonder-1.0: reviewer, coordinator, and verifier completed; read tools worked; the seeded division-instead-of-multiplication bug was verified with no incomplete passes. Exit 1 was expected for the request_changes result. No GitHub review was posted.
• GitHub test, ref-check, and review checks pass.

[github-actions]

🤖 AI code review

Decision: Ready for human review

Overall PR risk: Low. The change adds muse-wonder-1.0 support as an additive entry to META_MUSE_MODELS without modifying existing models. The affected surface is limited to buildOpencodeConfig runtime reads for this single model; all other configured models are unaffected. If metadata values are incorrect, only Wonder-specific runs would be impacted.

🔗 Review setup:

• This PR changes code the reviewer prompts cite (src). Check that the guidance quoting it is still correct.

No findings.

---

This review is advisory — it never blocks a merge and never auto-approves.