Skill retrieval 30 papers

Every note, newest paper first

1. Flat skill packs lift 20-book QA accuracy from 0.26 to 0.46 — a second routing level erases the gain

   2026-07-20 — Agent Skills packs — the folder-based standard for handing an agent on-demand expertise — expose only a short description until a task matches it, then load an indexed body, then the specific…

2. Task-decomposition reranking beats the best baseline 78.7 vs 73.1 on ALFWorld-unseen, using just 1.3 skills per task

   2026-07-07 — SkillReranker treats skill selection as a graph-matching problem, not a similarity search. It decomposes a task into an ordered sequence of subtasks and intermediate sub-states, then parses every…

3. A 3.9M-parameter skill sequencer closes 80% of the gap to hand-picked "gold" skill sets

   2026-06-30 — LLM agents built on skill libraries hit a bottleneck once the library grows: choosing what to load stops being a lookup problem and becomes a joint decision over subset, count, and order — three…

4. Compose agents from skills, not fixed roles: +2 points over the best topology-only baseline, only a 0.96-point dip when the skill library changes

   2026-06-18 — Existing graph-based multi-agent design treats agents as closed-set entities : fix a roster of agents, roles, or groups first, then optimize who talks to whom. SIGMA drops that assumption. Given a…

5. One Feedback Pass Takes Skill-Chain Decomposition From 51% to 68% Accuracy

   2026-06-16 — Compositional skill routing formalizes what happens when a query needs more than one skill: decompose it into atomic sub-tasks, retrieve a skill for each, then compose the results into an executable…

6. Verification-gated skills add up to 12 points on KernelBench — pull retrieval at inference and most of it vanishes

   2026-06-15 — daVinci-kernel splits CUDA/Triton kernel generation across three roles under one shared LLM backbone. A Selection Agent retrieves candidate optimization techniques through BM25 pre-filtering plus…

7. Keeping one skill per family drops harmful-sibling exposure from 69% to zero Bridge: retrieval × benchmarks

   2026-06-09 — A retriever can nail the capability family — the right general kind of skill for the job — and still hand back the wrong specific one. SkillResolve-Bench 1.0 names this failure same-capability…

8. A typed skill graph gives up just 3.5 retrieval points across a 10x library scale-up Bridge: retrieval × evolution

   2026-06-02 — SkillDAG models a large skill library as a typed directed graph instead of a flat embedding index. Five edge types — depends_on, specializes, composes_with, similar_to, conflicts_with — encode…

9. Even a Purpose-Built Reranker Nails the Full Skill Set Only 32% of the Time Bridge: retrieval × benchmarks

   2026-06-02 — LLM agents that compose multiple skills to solve a task depend on a retriever picking the right combination. That's a different problem from picking the right individual skill. A retriever can score…

10. Cutting 83% of skill calls raises task success 10.9 points

    2026-05-30 — SelSkill treats every point where an agent could call an agent skill as a binary decision — invoke or skip — instead of assuming relevance implies use. It flags candidate decision points with …

11. Dumping every skill into context scores worse than giving the agent none — SkillsInjector recovers it to 58.7%

    2026-05-28 — Skill injection is usually treated as a retrieval problem: rank candidates, keep the top K, paste their descriptions in unchanged. That static treatment turns out to be the bottleneck, not the skills…

12. One Rollout, 58.3% Success — Up to 61% Fewer Tokens Than Baseline Agents

    2026-05-24 — Web agents that click through visual interfaces usually get more capable by spending more at inference: sampling multiple rollouts, adding a verifier pass, or running an offline discovery phase…

13. Skill shadowing — not context bloat — drives up to 68% of a 202-skill library's 21% pass-rate drop

    2026-05-21 — A skill library doesn't have to get big before it starts working against you. Two Databricks researchers pin down exactly when: past the handful of skills a task actually needs, agents get…

14. One decay slope predicts both routing collapse and execution rescue — fixing it takes routing accuracy from 71.3% to 91.7% Bridge: retrieval × security

    2026-05-15 — Researchers stress-tested skill routing across 15 frontier LLMs and 1,141 real-world skills pulled from public Agent Skills repos, Claude Code, MCP servers, and community GitHub, organized into 14…

15. Dependency-ordered retrieval beats flat similarity search by 31 points on ALFWorld

    2026-05-12 — SkillGraph reframes an agent's skill library as a directed graph instead of a flat, similarity-ranked list. Each node is a skill distilled by a teacher model (OpenAI o3) from the agent's own…

16. Agent-skill catalogs already top 700,000 entries — curation hasn't caught up Bridge: retrieval × evolution

    2026-05-08 — Agent skills now have a formal definition: a skill is the tuple S = (M, R, C) — a root instruction document M, a set of auxiliary resources R (references, templates, scripts), and applicability…

17. Budget-aware skill packing takes deliverability from 9% to 100% for just 1.14 points of quality

    2026-05-08 — SkillSelect-Serve treats a reusable agent skill as a Skill Service — a structured record of what it does, its input/output shape, which tools it needs, how many tokens it costs to load, and a risk…

18. Role-labeled skill groups: reward 36%→49%, runtime cut 42%

    2026-05-07 — Retrieval for agent skill libraries has been treated as a search problem: find the k most relevant skills and hand them to the agent. GoSkills treats it as a presentation problem instead. It builds …

19. Splitting SKILL.md into three layers lifts retrieval 12%, risk detection 24% Bridge: retrieval × security

    2026-04-27 — LLM agent skills today live almost entirely as prose — SKILL.md -style files that bundle a skill's invocation rules, execution phases, and side effects into a single block of natural-language text.…

20. 0.000 to 0.805: a 42-skill library rescues a model that can't solve a single hard RTL problem alone

    2026-04-25 — LEGO breaks the digital front-end chip-design flow — spec generation, RTL coding, testbench generation, simulation, debug — into six fixed steps and represents every reusable capability within those…

21. Turning a corpus into a skill tree cuts RAG hallucination from 50% to 4.5%

    2026-04-16 — Corpus2Skill turns a document corpus into something an agent walks instead of queries. An offline compile pass embeds every document, clusters them hierarchically with K-Means, and has an LLM write a…

22. A skill graph beats context-stuffing: +25.55% reward, -56.72% tokens

    2026-04-07 — Graph-of-Skills (GoS) builds a typed skill graph offline over a local skill library, linking skills through four edge types: dependency, workflow, semantic, and alternative. Dependency edges come…

23. Cut Skill Bodies 39%, Task Success Rises 2.8%

    2026-03-31 — Agent skills are supposed to make agents cheaper and more capable, not the reverse. An audit of 55,315 publicly available skills finds the opposite happening at scale. 26.4% carry no routing…

24. 29 mined skills beat prompt-search code optimizers by up to 12.5 points — without running the code

    2026-03-29 — EffiSkill treats recurring slow-to-fast code rewrites as reusable agent skills instead of one-off prompts. Stage I mines Operator Skills (concrete transformation mechanics) and Meta Skills …

25. Skill Routing Loses 37-44 Points When It Can't See the Body

    2026-03-23 — LLM agent skill catalogs are heading into the same problem search ran into decades ago: too many candidates to hand a model all at once. The standard fix is progressive disclosure — a router picks…

26. 200k+ skills wired into one graph, +40% reward across three backbones Bridge: retrieval × benchmarks

    2026-02-26 — A nineteen-institution team spanning Zhejiang University, Alibaba, Tencent, Ant Group, and UCLA built SkillNet , infrastructure that creates, grades, and connects agent skills at scale instead of…

27. 478 GUI skills push GPT-5's desktop success rate up 15.6 points

    2026-01-28 — CUA-Skill turns human computer-use know-how into 478 atomic, parameterized skills across 17 Windows applications — File Explorer, Excel, Word, Chrome, VS Code, PowerPoint, and more. Each skill pairs…

28. One agent, one skill library: 54% cheaper than multi-agent — until it tops ~90 skills

    2026-01-08 — Multi-agent systems split a task across several specialized agents that talk to each other, and that coordination is expensive — every hand-off costs tokens and latency. Compile the same pipeline…

29. Retrieving the right skill lifts coding Pass@1 78%, reaching 84% of oracle-level performance

    2025-04-08 — SkillFlow treats skill selection as a search problem, not a context-stuffing problem. Given a task, it runs dense retrieval over roughly 36,000 community-contributed SKILL.md files scraped from…

30. Retrieving from a 210-skill library mines diamonds 2.5x faster than writing code on the fly

    2023-11-26 — STEVE is an embodied agent for Minecraft built from three linked modules: a vision encoder (EfficientFormer) that tokenizes what the agent sees, a LLaMA-2-13B instruction core split into four roles…