@prefix rdf: . @prefix rdfs: . @prefix owl: . @prefix xsd: . @prefix prov: . @prefix ws: . @prefix prop: . a ws:Subject ; rdfs:label "arxiv-2606-03056v1" . "SkillDAG: Self-Evolving Typed Skill Graphs for LLM Skill Selection at Scale" . [] a rdf:Statement, ws:AffirmedClaim ; rdf:subject ; rdf:predicate ; rdf:object "SkillDAG: Self-Evolving Typed Skill Graphs for LLM Skill Selection at Scale" ; prov:wasDerivedFrom . "As LLM agents adopt large skill libraries, selecting the right subset becomes a structural problem rather than a similarity-matching one: skills depend on, conflict with, specialize, or duplicate one another, a structure invisible to both full enumeration and embedding similarity. We present SkillDAG, which models inter-skill relationships as a typed directed graph and exposes it to an LLM agent as an inference-time, agent-callable structural retrieval interface, queried and evolved during execution rather than baked into a fixed retrieval pipeline: each search returns vector matches, typed-edge neighbors, and conflict signals, and a propose-then-commit protocol lets the agent register execution-backed edges so the graph accumulates structure across episodes. On ALFWorld and SkillsBench with MiniMax-M2.7, SkillDAG reaches 67.1% success and 27.3% reward, exceeding the strongest reported Graph-of-Skills baseline by +12.8 and +8.6 points; the advantage ports to gpt-5.2-codex, and intrinsic SkillsBench Ret@K rises from 65.5 to 78.2 under matched queries. These gains trace to isolable mechanisms: candidate ranking that stays robust as the pool grows 10x where a fixed seeding-diffusion pipeline degrades, and set-monotone online edits that enlarge ground-truth recall without evicting prior hits." . [] a rdf:Statement, ws:AffirmedClaim ; rdf:subject ; rdf:predicate ; rdf:object "As LLM agents adopt large skill libraries, selecting the right subset becomes a structural problem rather than a similarity-matching one: skills depend on, conflict with, specialize, or duplicate one another, a structure invisible to both full enumeration and embedding similarity. We present SkillDAG, which models inter-skill relationships as a typed directed graph and exposes it to an LLM agent as an inference-time, agent-callable structural retrieval interface, queried and evolved during execution rather than baked into a fixed retrieval pipeline: each search returns vector matches, typed-edge neighbors, and conflict signals, and a propose-then-commit protocol lets the agent register execution-backed edges so the graph accumulates structure across episodes. On ALFWorld and SkillsBench with MiniMax-M2.7, SkillDAG reaches 67.1% success and 27.3% reward, exceeding the strongest reported Graph-of-Skills baseline by +12.8 and +8.6 points; the advantage ports to gpt-5.2-codex, and intrinsic SkillsBench Ret@K rises from 65.5 to 78.2 under matched queries. These gains trace to isolable mechanisms: candidate ranking that stays robust as the pool grows 10x where a fixed seeding-diffusion pipeline degrades, and set-monotone online edits that enlarge ground-truth recall without evicting prior hits." ; prov:wasDerivedFrom . "http://arxiv.org/abs/2606.03056v1" . [] a rdf:Statement, ws:AffirmedClaim ; rdf:subject ; rdf:predicate ; rdf:object "http://arxiv.org/abs/2606.03056v1" ; prov:wasDerivedFrom . "2026-06-02T02:45:21Z" . [] a rdf:Statement, ws:AffirmedClaim ; rdf:subject ; rdf:predicate ; rdf:object "2026-06-02T02:45:21Z" ; prov:wasDerivedFrom . "Tong Bai" . [] a rdf:Statement, ws:AffirmedClaim ; rdf:subject ; rdf:predicate ; rdf:object "Tong Bai" ; prov:wasDerivedFrom . "Zhenglin Wan" . [] a rdf:Statement, ws:AffirmedClaim ; rdf:subject ; rdf:predicate ; rdf:object "Zhenglin Wan" ; prov:wasDerivedFrom . "Pengfei Zhou" . [] a rdf:Statement, ws:AffirmedClaim ; rdf:subject ; rdf:predicate ; rdf:object "Pengfei Zhou" ; prov:wasDerivedFrom . "Xingrui Yu" . [] a rdf:Statement, ws:AffirmedClaim ; rdf:subject ; rdf:predicate ; rdf:object "Xingrui Yu" ; prov:wasDerivedFrom . "Wangbo Zhao" . [] a rdf:Statement, ws:AffirmedClaim ; rdf:subject ; rdf:predicate ; rdf:object "Wangbo Zhao" ; prov:wasDerivedFrom . "Yang You" . [] a rdf:Statement, ws:AffirmedClaim ; rdf:subject ; rdf:predicate ; rdf:object "Yang You" ; prov:wasDerivedFrom . "Ivor W. Tsang" . [] a rdf:Statement, ws:AffirmedClaim ; rdf:subject ; rdf:predicate ; rdf:object "Ivor W. Tsang" ; prov:wasDerivedFrom .