Reading skill routing out of the model's own forward pass beats retrieval pipelines carrying 16B extra parameters.

Deployed harnesses preload every skill's metadata into context, which disperses attention and caps library size; retrieval pipelines move selection out of context and out of the agent's capability at the same time (arXiv 2609.15982). Gavel trains two linear maps to read mid-layer states, scores the full library against per-skill banks built in one forward pass at installation, then resumes the shortlisted skills' forward passes for a yes/no verdict fused as a product of experts. On Qwen3-32B it beats progressive disclosure and retrieve-and-rerank by up to 13.4 points on written tasks and 21.9 points when the need for a skill arises mid-rollout, with no skill text in context at all.