{
  "schema_version": "1.1",
  "id": "s13:https://arxiv.org/abs/2609.15982v1",
  "slug": "2609-15982v1-1b049zj",
  "url": "https://feed7.dev/p/2609-15982v1-1b049zj",
  "title": "The Router Within: Eliciting Native Skill Routing from a Frozen LLM",
  "why_included": "Gavel routes skills from a frozen model’s internal states without loading skill text into context. It suggests large skill libraries may need model-native selection rather than metadata-heavy prompts.",
  "summary": "Gavel reads skill-routing signals from a frozen LLM using **two trained linear maps**. A glance scores the full library from mid-layer states and prebuilt skill banks; a verdict resumes shortlisted forward passes and combines likelihood with a yes/no judgment.",
  "practical_implication": "Harness authors should benchmark routing separately from skill execution, especially when skills become relevant mid-run. This design keeps skill descriptions out of the active context and avoids external rerankers with **1.2B–16B parameters**.",
  "agent_context": "Gavel reads skill-routing signals from a frozen LLM using **two trained linear maps**. A glance scores the full library from mid-layer states and prebuilt skill banks; a verdict resumes shortlisted forward passes and combines likelihood with a yes/no judgment.\n\nHarness authors should benchmark routing separately from skill execution, especially when skills become relevant mid-run. This design keeps skill descriptions out of the active context and avoids external rerankers with **1.2B–16B parameters**.\n\nOn Qwen3-32B, Gavel leads compared pipelines by **up to 13.4 points** on written tasks and **up to 21.9 points** for mid-rollout skill needs. The claims span three public benchmarks plus a new simulated-trajectory benchmark, so production generality remains open.",
  "source": {
    "name": "arXiv",
    "url": "https://arxiv.org/abs/2609.15982v1",
    "published_at": "2026-09-14T17:58:27.000Z"
  },
  "source_class": "blog_post",
  "content_type": "Paper",
  "layer": "agent",
  "domains": [
    "coding"
  ],
  "topics": [
    "skills",
    "tool-use",
    "context-engineering"
  ],
  "verification": {
    "status": "needs_review",
    "label": "Needs Review",
    "method": "unverified",
    "verified_at": null
  },
  "uncertainty": [
    "On Qwen3-32B, Gavel leads compared pipelines by **up to 13.4 points** on written tasks and **up to 21.9 points** for mid-rollout skill needs. The claims span three public benchmarks plus a new simulated-trajectory benchmark, so production generality remains open."
  ],
  "connected_context": {
    "meaning": "Gavel shifts progressive skill discovery inside the frozen model: compact learned probes shortlist skills without placing the entire catalog in context or adding a large reranker, including when relevance appears mid-rollout. It strengthens the case for evaluating routing as its own subsystem, while leaving governance, execution safety, model-change reevaluation, and production transfer outside the reported routing gains.",
    "corpus_size": 778,
    "generated_at": "2026-09-15T10:07:01.848Z",
    "connections": [
      {
        "title": "Skills are new features: Building Skill-Centric Harness — Yogendra Miraje, FactSet",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=7jjudsEhBtM",
        "feed7_url": "https://feed7.dev/p/skills-are-new-features-building-skill-centric-harness-yogendra-miraje-f-0lp4c7o",
        "reason": "Gavel offers a concrete routing mechanism for the skill-centric harness, while FactSet’s account supplies the surrounding requirements—versioning, model-specific evaluation, access control, and governance—that routing accuracy alone does not cover."
      },
      {
        "title": "500 Skills, Zero Fine-Tuning: LinkedIn's Playbook for AI Agents — Ajay Prakash, LinkedIn",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=9wZpvF3QleU",
        "feed7_url": "https://feed7.dev/p/500-skills-zero-fine-tuning-linkedin-s-playbook-for-ai-agents-ajay-praka-1duois1",
        "reason": "LinkedIn demonstrates progressive discovery through search and meta-tools at large catalog scale; Gavel proposes a contrasting native-model shortlist-and-verdict pipeline that keeps skill descriptions out of active context."
      },
      {
        "title": "Agents Without Code: Skills, YAML, and Filesystems Replaced Python — Philipp Schmid, Google DeepMind",
        "source_name": "AI Engineer",
        "source_url": "https://www.youtube.com/watch?v=fjF8EKnxKCU",
        "feed7_url": "https://feed7.dev/p/agents-without-code-skills-yaml-and-filesystems-replaced-python-philipp-0t7a4s8",
        "reason": "The low-code agent pattern reduces bespoke orchestration but leaves evaluation to builders; Gavel makes routing evaluation especially necessary because skill selection is learned separately from execution."
      },
      {
        "title": "zhaoxuya520/reverse-skill",
        "source_name": "GitHub",
        "source_url": "https://github.com/zhaoxuya520/reverse-skill",
        "feed7_url": "https://feed7.dev/p/reverse-skill-1e4jlfw",
        "reason": "The security router shows why selection must be paired with authorization and evidence stages; Gavel may improve which skill is chosen, but its reported design does not replace external controls around risky execution."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-09-14T17:58:27.000Z",
  "modified_at": "2026-09-14T17:58:27.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/2609-15982v1-1b049zj",
    "json": "https://feed7.dev/p/2609-15982v1-1b049zj.json",
    "markdown": "https://feed7.dev/p/2609-15982v1-1b049zj.md"
  }
}