feat(models): add grok-composer-2.5-fast to xAI model list

[lidge-jun]

Summary

• Adds x-ai/grok-composer-2.5-fast to the xAI model list in hermes_cli/models.py
• This model is live on xAI /v1/chat/completions with reasoning_content support
• It is Grok Build's composer model for agentic code composition
• Not yet listed in /v1/models or /v1/language-models but responds to direct calls

Test plan

• Verify hermes auth add xai-oauth → hermes run --model x-ai/grok-composer-2.5-fast works
• Confirm model appears in hermes models output

🤖 Generated with Claude Code

[lidge-jun]

CI note: test (5) failure is a pre-existing flaky shard — this PR only adds one line to hermes_cli/models.py and does not touch test code. The other 22/23 checks pass.

I've verified grok-composer-2.5-fast is live on xAI /v1/chat/completions with tool_use and reasoning_content support.

[lidge-jun]

Update: CI fix pushed + behavior proof

CI fix (second commit)

The test (5) failure was our fault, not a pre-existing flaky shard:

FAILED tests/hermes_cli/test_model_catalog.py::TestManifestMatchesInRepoLists::test_in_repo_lists_match_manifest
AssertionError: website/static/api/model-catalog.json is out of sync with _PROVIDER_MODELS['nous']

Adding x-ai/grok-composer-2.5-fast to models.py without regenerating model-catalog.json caused the manifest sync check to fail. Fixed by running python scripts/build_model_catalog.py and committing the regenerated JSON.

Second commit: f9c29f4 — rebuilds model-catalog.json with the new entry.

Live behavior proof

Tested on macOS with progrok OAuth session, June 2, 2026:

Chat completion with reasoning:

model: grok-composer-2.5-fast
reasoning_content: "Okay, the user wants me to write a fibonacci function..."
content: "def fibonacci(n: int) -> int: ..."
usage: 1190 tokens, cost_in_usd_ticks: 23566000
finish_reason: stop

Function calling (tool_use):

{
  "tool_calls": [{
    "id": "call-...-composer_call_EYkpf",
    "function": {
      "name": "generate_image",
      "arguments": "{\"prompt\":\"A breathtaking sunset...\"}"
    }
  }],
  "finish_reason": "tool_calls"
}

Capabilities verified:

• ✅ /v1/chat/completions — HTTP 200
• ✅ reasoning_content — returned in message
• ✅ tool_use with forced tool_choice — returns tool_calls
• ✅ System prompt, multi-turn, temperature control
• ❌ Not listed in /v1/models (hidden model)
• ❌ stop sequences — HTTP 400 (unsupported)
• 💰 Pricing: ~$1.00/M input, ~$2.00/M output

Adoption

Already live in:

• progrok — default planner model
• ima2-gen — switched from grok-4.3 to composer for image/video planning
• openclaw PR #89190
• opencode PR #30280