Plugin SDK: supportsAdaptiveThinking allowlist missing claude-opus-4-7

This issue is being filed as part of a batch of related upstream submissions discovered during architectural review of OpenClaw deployment. Filed locally 2026-04-23; submitted upstream 2026-04-27. See "Related issues (batch)" section at end for batch context.

Suggested labels (Mike to choose from labels actually available in target repo at submission time): bug, severity:high, plugin-sdk, transport:anthropic, model:claude-opus-4-7

Affected versions: v2026.4.15 (confirmed); earlier versions likely (14 fires since Apr 20)
Severity: HIGH — every Opus-4.7 invocation with thinking enabled fails immediately
Scope: All anthropic provider calls routing through OpenClaw's vertex-stream transport for claude-opus-4-7 when options.thinkingEnabled === true

Upstream project: OpenClaw plugin SDK + anthropic-vertex-stream transport

Diagnosis

File: dist/anthropic-vertex-stream-BpkPWKP9.js, lines 5064-5071

if (model.reasoning) {
    if (options?.thinkingEnabled) 
        if (supportsAdaptiveThinking(model.id)) {
            params.thinking = { type: "adaptive" };
            if (options.effort) params.output_config = { effort: options.effort };
        } else params.thinking = {                    // <-- hits this branch on Opus 4.7
            type: "enabled",                           // <-- wrong schema for Opus 4.7
            budget_tokens: options.thinkingBudgetTokens || 1024
        };
    else if (options?.thinkingEnabled === false) params.thinking = { type: "disabled" };
}

supportsAdaptiveThinking(model.id) currently returns false for claude-opus-4-7. Opus 4.7 requires the adaptive schema and rejects the legacy type: "enabled" schema with HTTP 400:

"thinking.type.enabled" is not supported for this model. Use "thinking.type.adaptive" and "output_config.effort" to control thinking behavior.

Evidence

Gateway journal counts for thinking.type.enabled is not supported:

• Today (2026-04-23): 9 fires across 2 main-agent runs + 5 subagent spawns (mccoy×2, athena, cathie, zell) + FT-4 probe
• Since 2026-04-20: 14 fires total
• 100% of fires involve model=claude-opus-4-7 provider=anthropic

Fix

Add claude-opus-4-7 to the supportsAdaptiveThinking model allowlist, OR audit all Opus-4.7 model descriptors to ensure they route through the adaptive branch. The fix lives in whatever supportsAdaptiveThinking function body references; grep for that function name.

Workaround

Disable thinking on Opus-4.7 invocations where thinking isn't essential: thinking: 'off' / thinkingEnabled: false in the spawn options. FT-4 v3 probe uses this workaround.

Related

• Same pattern expected for any future Opus model variants until allowlist updated.
• Affected agents today: main-agent (2), mccoy (2 subagent spawns), athena, cathie, zell, FT-4 probe session.
• This was the secondary cause (first was identity-isolation) of FT-4 v2 probe failure at 19:55:05.

Request for upstream review

Please consider:

1. Adding claude-opus-4-7 to the supportsAdaptiveThinking model allowlist.
2. Auditing all Opus-4.7 model descriptors to ensure they route through the adaptive branch.
3. Evaluating whether future Opus model variants will require the same allowlist addition (suggesting a defaulting pattern that catches new models automatically).

Related issues (batch)

This issue is filed as part of a batch of related upstream submissions:

• [anthropics/claude-code#53710] F12-A: Glob tool false-negative on recently-created files — Anthropic Claude Code repo
• [#72541] F11-C: Gateway completeness check false-negative on pure-relay agents — OpenClaw repo
• [#72540] Opus 4.7: supportsAdaptiveThinking allowlist missing claude-opus-4-7 — OpenClaw repo
• [#72539] Sub-B-FAR: Plugin SDK MessagePreSent outbound hook missing — OpenClaw repo