[edverma]

Summary

Adds Muse Wonder + Wonder Full (muse-wonder-1.0[-contributor], muse-wonder-full-1.0[-contributor]) to the muse-code provider.

Live GET https://api.meta.ai/v1/models now returns 12 ids including the 4 Wonder variants. The muse-spark--only allowlist in src/catalog.ts dropped them from live discovery, and the static baseline + openclaw.plugin.json only listed 5x Spark.

Changes:

• src/catalog.ts: CHAT_FAMILY_PREFIX → CHAT_FAMILY_PREFIXES = ["muse-spark-", "muse-wonder-"]
• src/baseline.models.ts + openclaw.plugin.json + rebuilt dist/: 5 → 9 models (context 1048576, maxTokens 131072, text+image, 0 cost — same as Spark; live rows carry no context_length)
• index.ts description/comments, README.md table, package.json 0.2.0 → 0.3.0 + keyword
• test/catalog.test.ts, test/modelMapping.test.ts updated

How verified

• npm run build passes and committed dist/ is in sync
• npm test passes (39 tests, 3 files)
• Added/updated tests for the change
• Live on vm-metis-openclaw, OpenClaw 2026.9.9: models list shows 9 muse-code models; agent --model muse-code/muse-wonder-1.0 → WONDER-OK; muse-wonder-full-1.0 → WONDER-FULL-OK

Notes

• No new dependencies. No secrets/API keys committed.
• Default stays muse-spark-1.3 (never contributor).

[TheStreamCode]

Thank you: the live catalog observation (12 ids including the Wonder variants) is exactly the evidence needed. I am closing in favor of a different direction: discovery is moving to a fully dynamic list driven by the live endpoint (minus known non-chat families), so no per-model code, baseline, or manifest edits are needed going forward. Your report confirmed the Wonder IDs are live. Please test the dynamic release when it lands.

[TheStreamCode]

v0.3.0 is out with fully dynamic discovery: every id from the live catalog is served except the verified non-chat families (sam-, muse-image-, muse-voice-*), so the 12-id response you captured — including the 4 Wonder variants — now flows straight through, with the static baseline kept as offline fallback only. Your live verification was the key evidence for this direction. If you get a chance, re-running your checks (models list, agent --model muse-code/muse-wonder-1.0) against v0.3.0 would be valuable confirmation — and thanks again.