Summary

An agent session accumulated 237,817 tokens (exceeding Claude's 200k limit) after calling gateway(action: "config.schema"), which returned 394,396 characters (~100k tokens) of JSON Schema data.

This caused repeated embedded run timeout errors as every subsequent API call failed with prompt is too long.

Root Cause

When an agent wanted to check logging format options, it called:

{"action": "config.schema"}

The agent only needed to understand logging configuration, but received the entire OpenClaw configuration schema - including all channels, plugins, UI hints, and field definitions.

Evidence

From session 84018b53-6210-4868-a7ed-62c6e83d42cb:

Error message:

prompt is too long: 237817 tokens > 200000 maximum

Suggested Fixes

Option A: Truncate or summarize large toolResults

Add a size limit to toolResult content before adding to session history. For schema-like responses, return a summary instead of full content.

Option B: Add scope parameter to config.schema

// Instead of returning everything:
gateway({ action: "config.schema" })

// Allow scoped queries:
gateway({ action: "config.schema", scope: "logging" })
gateway({ action: "config.schema", scope: "channels.telegram" })

Option C: Session history auto-compression

When session approaches token limit, automatically compress/summarize older tool results.

Option D: Warn agent about large responses

Before executing config.schema, warn that it returns a large response and suggest alternatives like reading docs or using config.get.

Related Code

• src/agents/tools/gateway-tool.ts:171-174 - tool entry point
• src/gateway/server-methods/config.ts:110-151 - handler
• src/config/schema.ts:1081-1103 - buildConfigSchema() generates full schema

Impact

• 10 consecutive embedded run timeout errors over 7.5 hours
• Agent completely blocked until session is manually cleared
• Wastes API quota on failed requests

Discovered while investigating repeated timeout errors in gateway logs.