feat(xai): add grok-composer-2.5-fast model

[lidge-jun]

Summary

• Adds grok-composer-2.5-fast to xAI model catalog in extensions/xai/model-definitions.ts
• Adds grok-composer prefix to modern model list in extensions/xai/provider-models.ts
• Model is live on xAI /v1/chat/completions with reasoning_content support
• Grok Build's composer model for agentic code composition

Test plan

• Verify model appears in openclaw models with xAI provider configured
• Verify chat completion works with grok-composer-2.5-fast

🤖 Generated with Claude Code

[clawsweeper]

Codex review: needs real behavior proof before merge. Reviewed June 29, 2026, 8:05 PM ET / 00:05 UTC.

Summary
The PR adds grok-composer-2.5-fast to the xAI static catalog and treats grok-composer as a modern xAI model prefix.

PR surface: Source +9. Total +9 across 2 files.

Reproducibility: yes. for the review finding: source inspection shows xAI catalog rows are filtered by XAI_SELECTABLE_MODEL_IDS and onboarding prunes non-selectable built-in ids. I did not run the CLI because this is a read-only review.

Review metrics: 1 noteworthy metric.

• xAI static catalog exposure: 1 catalog entry added, 0 selectable ids added. A non-selectable built-in catalog row affects retired-model pruning without making the model visible in the static picker path.

Merge readiness
Overall: 🦪 silver shellfish
Proof: 🦪 silver shellfish
Patch quality: 🦪 silver shellfish
Result: blocked until stronger real behavior proof is added.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Rank-up moves:

• [P1] Add redacted OpenClaw after-patch proof for model listing or a configured composer chat run.
• Choose static selectable support versus OAuth-live-only support, then align tests and docs with that contract.

Proof guidance:

• [P1] Needs stronger real behavior proof before merge: The posted live output exercises direct xAI/progrok calls, but not the changed OpenClaw model listing, onboarding, selection, or chat path after this patch. After adding proof, update the PR body; ClawSweeper should re-review automatically. If it does not, the PR author or someone with repository write access can comment @clawsweeper re-review.

Risk before merge

• [P1] Merging as-is can make xAI onboarding/configure treat an existing manually added composer model as a retired built-in and remove it from provider config.
• [P1] The PR creates a second source of truth for a hidden upstream xAI model while current main already exposes composer through OAuth live discovery.
• [P1] The posted proof shows direct provider behavior, not the changed OpenClaw model listing, selection, onboarding, or chat path after this patch.

Maintainer options:

1. Choose the composer support contract first (recommended)
   Maintainers should decide whether composer is a selectable static/API-key model or remains OAuth-live-discovered before accepting static catalog changes.
2. Make static support coherent
   If static support is intended, add composer to the selectable contract or otherwise prevent prune behavior, then cover listing/onboarding with focused xAI tests and OpenClaw proof.
3. Close in favor of live OAuth discovery
   If the merged live catalog path is the intended support route, close this PR rather than adding a parallel hidden-model catalog source.

Next step before merge

• [P1] The remaining blocker is a maintainer xAI hidden-model support decision plus contributor OpenClaw proof, not a safe automated repair.

Security
Cleared: The diff only changes xAI model metadata and prefix classification; it does not add dependencies, workflows, scripts, credential handling, or supply-chain-sensitive code.

Review findings

• [P1] Classify composer consistently before publishing it — extensions/xai/model-definitions.ts:198

Review details

Best possible solution:

Decide the xAI composer support contract, then either close in favor of OAuth live discovery or land coherent static/API-key support with selectable catalog behavior, focused tests, and OpenClaw proof.

Do we have a high-confidence way to reproduce the issue?

Yes for the review finding: source inspection shows xAI catalog rows are filtered by XAI_SELECTABLE_MODEL_IDS and onboarding prunes non-selectable built-in ids. I did not run the CLI because this is a read-only review.

Is this the best way to solve the issue?

No, not yet: the patch is a plausible static-support slice, but current main already supports composer through xAI OAuth live discovery and the PR does not settle whether static/API-key hidden-model support should be public.

Full review comments:

• [P1] Classify composer consistently before publishing it — extensions/xai/model-definitions.ts:198
  Adding this id to XAI_MODEL_CATALOG without adding it to XAI_SELECTABLE_MODEL_IDS makes it a non-selectable built-in. That keeps it out of the static picker path while isRetiredXaiBuiltinModelId can prune a manually configured composer row during onboarding; either make static support fully selectable and tested, or rely on the existing live OAuth discovery path.
  Confidence: 0.88

Overall correctness: patch is incorrect
Overall confidence: 0.88

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 18b2ff683fd3.

Label changes

Label justifications:

• P2: This is a narrow xAI provider model-catalog feature with limited blast radius, but it affects model selection and provider routing.
• merge-risk: 🚨 compatibility: The PR can classify a manually configured composer model as a retired built-in and remove it during xAI onboarding.
• merge-risk: 🚨 auth-provider: The PR changes xAI model exposure for a hidden upstream model whose supported auth path still needs a maintainer decision.
• rating: 🦪 silver shellfish: Overall readiness is 🦪 silver shellfish; proof is 🦪 silver shellfish and patch quality is 🦪 silver shellfish.
• status: 📣 needs proof: The PR needs real behavior proof before ClawSweeper can clear the contributor ask. Needs stronger real behavior proof before merge: The posted live output exercises direct xAI/progrok calls, but not the changed OpenClaw model listing, onboarding, selection, or chat path after this patch. After adding proof, update the PR body; ClawSweeper should re-review automatically. If it does not, the PR author or someone with repository write access can comment @clawsweeper re-review.

Evidence reviewed

PR surface:

Source +9. Total +9 across 2 files.

View PR surface stats

Area	Files	Added	Removed	Net
Source	2	10	1	+9
Tests	0	0	0	0
Docs	0	0	0	0
Config	0	0	0	0
Generated	0	0	0	0
Other	0	0	0	0
Total	2	10	1	+9

What I checked:

• PR diff adds static composer metadata: The branch adds a grok-composer-2.5-fast row to XAI_MODEL_CATALOG and adds grok-composer to XAI_MODERN_MODEL_PREFIXES. (extensions/xai/model-definitions.ts:198, a4a7d2af5d29)
• Static catalog only publishes selectable ids: Current main builds xAI static model rows by filtering XAI_MODEL_CATALOG through XAI_SELECTABLE_MODEL_IDS; the PR adds no selectable id for composer. (extensions/xai/model-definitions.ts:260, 18b2ff683fd3)
• Non-selectable built-ins are pruned during onboarding: xAI onboarding filters configured xAI provider rows through isRetiredXaiBuiltinModelId, so adding a catalog row without making it selectable can remove an existing manually configured composer row. (extensions/xai/onboard.ts:29, 18b2ff683fd3)
• Current main has OAuth live composer coverage: The xAI runtime catalog maps OAuth /models rows from the Grok proxy, includes grok-composer-2.5-fast in the test fixture, and marks composer reasoning for replay support. (extensions/xai/index.test.ts:150, 18b2ff683fd3)
• Merged overlapping live-catalog work: Merged PR feat: add live provider model catalog helper #90029 added the live provider catalog helper and xAI OAuth discovery; its posted proof includes xAI OAuth rows containing composer, but it does not settle static/API-key hidden-model support. (extensions/xai/provider-catalog.ts:175, 57e0bdaabe0a)
• Posted proof does not exercise OpenClaw after the patch: The contributor posted direct xAI/progrok output for chat completions, reasoning content, and tool calls, but not a patched OpenClaw model listing, selection, onboarding, or chat run. (a4a7d2af5d29)

Likely related people:

• fuller-stack-dev: Authored the merged live provider catalog work that added xAI OAuth discovery and current composer coverage in tests and proof. (role: live provider catalog feature contributor; confidence: high; commits: 57e0bdaabe0a, 75afe981df63, 94f215846f91; files: extensions/xai/provider-catalog.ts, extensions/xai/index.test.ts, src/plugin-sdk/provider-catalog-live-runtime.ts)
• steipete: Feature history shows the original xAI provider integration and follow-up live catalog refactors touching the same provider/catalog boundary. (role: xAI provider and catalog contributor; confidence: medium; commits: a8907d80ddac, aeeca61712a1, b398919e76d3; files: extensions/xai/index.ts, extensions/xai/model-definitions.ts, extensions/xai/provider-models.ts)
• geraint0923: Authored the merged xAI reasoning PR that changed replay behavior for reasoning-capable xAI models, with composer reasoning adjusted in that PR's history. (role: recent adjacent xAI reasoning contributor; confidence: medium; commits: e8a31ddbcedc, 76a26360036f; files: extensions/xai/provider-catalog.ts, extensions/xai/runtime-model-compat.ts, extensions/xai/stream.ts)

What the crustacean ranks mean

• 🦀 challenger crab: rare, exceptional readiness with strong proof, clean implementation, and convincing validation.
• 🦞 diamond lobster: very strong readiness with only minor maintainer review expected.
• 🐚 platinum hermit: good normal PR, likely mergeable with ordinary maintainer review.
• 🦐 gold shrimp: useful signal, but proof or patch confidence is still limited.
• 🦪 silver shellfish: thin signal; proof, validation, or implementation needs work.
• 🧂 unranked krab: not merge-ready because proof is missing/unusable or there are serious correctness or safety concerns.
• 🌊 off-meta tidepool: rating does not apply to this item.

Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

How this review workflow works

• ClawSweeper keeps one durable marker-backed review comment per issue or PR.
• Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
• A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
• PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
• Maintainers can also comment @clawsweeper review to request a fresh review only.
• Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
• Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
• Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.

Review history (1 earlier review cycle)

• reviewed 2026-06-21T10:17:00.067Z sha a4a7d2a :: needs real behavior proof before merge. :: [P1] Classify composer consistently before publishing it

[lidge-jun]

CI note: "Real behavior proof" failures require live xAI credentials which are not available in fork CI. This PR only adds model definitions — no runtime behavior changes.

Verified: grok-composer-2.5-fast responds on xAI /v1/chat/completions with tool_use (function calling) and reasoning_content support. Pricing estimated ~$1.00/$2.00 per 1M tokens.

[lidge-jun]

Real Behavior Proof — grok-composer-2.5-fast

Captured on macOS, June 2, 2026, using progrok OAuth session and direct xAI API calls.

Proof 1: progrok models command (shows [composer] tag detection)

Available Grok models:
  grok-4.20-0309-non-reasoning [reasoning]
  grok-4.20-0309-reasoning [reasoning]
  grok-4.20-multi-agent-0309
  grok-4.3
  grok-build-0.1 [code]
  grok-imagine-image [image]
  grok-imagine-image-quality [image]
  grok-imagine-video [video]
  grok-imagine-video-1.5-preview [video]

Note: grok-composer-2.5-fast does not appear in /v1/models — it's a hidden model accessible via direct /v1/chat/completions calls.

Proof 2: Chat completion with reasoning_content

model: grok-composer-2.5-fast
reasoning_content: "Okay, the user wants me to write a fibonacci function
  in Python with type hints. Let me start by recalling what the Fibonacci
  sequence is..."
content: "Here's a Python function that calculates the nth Fibonacci number
  using an efficient iterative approach with type hints:
  def fibonacci(n: int) -> int: ..."
usage: {"prompt_tokens":21,"completion_tokens":1169,"total_tokens":1190,
  "cost_in_usd_ticks":23566000}
finish_reason: stop

Proof 3: tool_use (function calling) — forced tool_choice

{
  "tool_calls": [{
    "id": "call-f02d2838-5196-4dda-8df1-c16d9548e542-composer_call_EYkpf",
    "function": {
      "name": "generate_image",
      "arguments": "{\"prompt\":\"A breathtaking sunset over the ocean...\"}"
    },
    "type": "function"
  }],
  "finish_reason": "tool_calls"
}

Summary

Capability	Result
/v1/chat/completions	✅ HTTP 200
reasoning_content	✅ Returned in response
tool_use with forced tool_choice	✅ Returns tool_calls
System prompt	✅ Works
Multi-turn	✅ Works
Pricing	~$1.00/M in, ~$2.00/M out (from cost_in_usd_ticks)
Listed in /v1/models	❌ Hidden model
stop sequences	❌ HTTP 400 (unsupported)

[lidge-jun]

CI update (post-rebase)

Two failures, both unrelated to this PR's changes:

1. Real behavior proof — Requires live xAI credentials not available in fork CI. Expected for external contributors.

2. checks-node-agentic-agents-core-runner — session-write-lock.test.ts flaky failure:

   SessionWriteLockStaleError: session file lock stale (missing-pid, invalid-createdAt)
   

   This test creates temp lock files and races with PID/timestamp validation — a timing-sensitive flaky test. This PR only adds model entries to model-definitions.ts and a prefix to provider-models.ts; no lock or session code was touched.

This PR's actual changes (2 files, docs-level model catalog additions) have no test surface to fail.

[steipete]

Thanks @lidge-jun for validating Composer when it launched.

Closing this as superseded by #90029 (landed commit) and #95686. Current main uses xAI’s provider-owned OAuth catalog: it calls the Grok subscription proxy’s /v1/models endpoint and exposes grok-composer-2.5-fast only when the authenticated account advertises it, including reasoning replay support.

That matches xAI’s current contract. Composer 2.5 is announced as a Grok Build subscription model, while the public api.x.ai catalog is scoped to the authenticating API key. A fresh authenticated API-key probe returned HTTP 200 from /v1/models without Composer, and both the exact model lookup and a Responses request returned HTTP 404.

The proposed static row also remains outside XAI_SELECTABLE_MODEL_IDS, so it would not appear in the static picker or API-key catalog; the prefix addition only changes live-smoke classification. The supported OpenClaw path is openclaw models auth login --provider xai --method oauth, followed by selecting Composer when xAI advertises it.

If xAI later publishes Composer as a supported api.x.ai model, please open a fresh issue with current official docs and /v1/models evidence so we can add it coherently. Thanks again.