Summary

When using the built-in coding tools profile with a non-OpenAI provider (e.g. Anthropic), the gateway logs a warning on every session init:

[tools] tools.profile (coding) allowlist contains unknown entries (apply_patch).
These entries won't match any tool unless the plugin is enabled.

This fires repeatedly — on every session start, reset, and tool policy evaluation.

Root Cause

The coding profile's allowlist includes apply_patch (which is an OpenAI-specific tool for multi-file patches). When the active provider is Anthropic/Claude, apply_patch is never registered as a tool, so stripPluginOnlyAllowlist() in tool-policy-pipeline.ts flags it as an unknown entry.

The tool IS in DEFAULT_TOOL_ALLOW and TRUSTED_TOOL_RESULT_MEDIA, so it's intentionally part of the ecosystem — it's just provider-specific.

Impact

• Warning noise in logs on every session lifecycle event
• Confusing for users who haven't configured anything custom — the warning suggests misconfiguration but it's the built-in profile
• Makes it harder to spot real warnings in log output

Expected Behavior

Provider-specific tools in built-in profiles should either:

1. Be silently ignored when the provider doesn't support them (no warning)
2. Be conditionally included in the profile based on the active provider
3. At minimum, the warning should be downgraded to debug level for built-in profile entries

Reproduction

1. Set tools.profile: "coding" in openclaw.json (or use it as default)
2. Use any non-OpenAI provider (e.g. anthropic/claude-sonnet-4-6)
3. Start a session
4. Check gateway.err.log — warning appears

Environment

• OpenClaw 2026.3.7
• Provider: anthropic/claude-opus-4-6
• Config: tools.profile: "coding", tools.exec.security: "full"