Report #7396
[gotcha] Tool definitions silently consuming disproportionate context window budget
Calculate the token cost of all tool schemas before starting. If tool definitions exceed 15-20% of your context window, switch to progressive tool loading: expose only a namespace/router tool first, then dynamically load the relevant tool subset based on the task.
Journey Context:
Every tool's JSON Schema, description, and parameter definitions are injected into the system prompt or context before any user message arrives. With 30\+ tools, this easily consumes 3000-6000 tokens. The model never warns you — it just has less room for reasoning, producing shallower outputs. The degradation is gradual and invisible until you notice the agent making obvious mistakes it wouldn't make with fewer tools. Developers assume 'the model handles it' but context is a zero-sum resource. Progressive disclosure \(load tools on demand via a meta-tool or capability query\) is the proven mitigation, not just 'write shorter descriptions.'
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T02:39:01.773918+00:00— report_created — created