{"type":"post","slug":"skill-router-must-say-no","title":"A Skill Router Must Be Able to Say No","description":"A practical routing contract for loading agent skills only when measured task evidence predicts a net benefit.","retrieval_nugget":"A useful skill router must be allowed to return no skill when evidence does not predict a net gain. These results come from bounded benchmark harnesses and do not establish one universal skill policy across models, repositories, or production environments.","published_at":"2026-08-27","updated_at":"2026-08-27","record_date":"2026-08-27","date_kind":"published_at","topics":["agents","architecture"],"entities":[],"source_urls":["https://arxiv.org/abs/2608.14036","https://arxiv.org/abs/2608.23067","https://arxiv.org/abs/2608.19880"],"source_format":"research synthesis","editorial_timing":{"lane":"regular_hourly","scheduled_at":"2026-08-30T17:00:00+03:00","real_news_delta":"Recent studies disagree mainly because they test different exposure regimes, making unconditional semantic matching an unsafe default."},"schema_version":"newruntime-agent-readable-v0.2","stable_id":"post:skill-router-must-say-no","status":"published","visuals":[],"editorial_provenance":{"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:49c9251bced9bc799b5b8f95be2c0d288610137c3f317c452a9af925141a9d92","reviewed_at":"2026-08-27T16:57:21Z","source_evidence_count":1,"verified_claim_count":2,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":2,"implication_count":1,"watch_condition_count":1,"related_record_count":1},"analysis":{"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"A useful skill router must be allowed to return no skill when evidence does not predict a net gain.","observed_facts":[{"text":"Demystifying Agent Skills reports a 6.06 percentage-point advantage for skill artifacts over workflow memory built from the same experience.","source_urls":["https://arxiv.org/abs/2608.14036"]},{"text":"Its trajectory analysis attributes 65.7 percent of observed skill mechanisms to procedural anchoring and only 4.5 percent to explicit knowledge injection.","source_urls":["https://arxiv.org/abs/2608.14036"]}],"mechanism":"The router should compare task evidence with a no-skill baseline, load a narrow procedure only for a diagnosed execution gap, and measure the resulting failure type and token cost.","why_now":"Recent studies disagree mainly because they test different exposure regimes, making unconditional semantic matching an unsafe default.","implications":["Treat each skill as a versioned policy patch with a trigger, an exit condition, and task-level evaluation against the agent's native behavior."],"evidence_boundary":"These results come from bounded benchmark harnesses and do not establish one universal skill policy across models, repositories, or production environments.","watch_conditions":["Revise the router when a skill's measured gain disappears after a model, harness, dependency, or task distribution changes."],"related_records":[{"url":"https://newruntime.com/patterns/skills-become-portable-capability-layer","relation":"This note turns the existing capability-layer thesis into a conditional loading rule."}]},"routes":{"html":"https://newruntime.com/posts/skill-router-must-say-no/","markdown":"https://newruntime.com/posts/skill-router-must-say-no.md","json":"https://newruntime.com/posts/skill-router-must-say-no.json"}}