Reusable skill libraries allow large language model (LLM) agents to reuse procedural knowledge across tasks, but they also turn memory access into a challenging retrieval problem. Full-library prompting preserves coverage at high context cost, vector retrieval returns compact neighborhoods but treats skills as independent text, and graph-based retrieval can recover workflow context only when the edges that carry relevance are reliable. We propose CaSKG, a counterfactual-causal skill graph framework that calibrates procedural relations before retrieval. CaSKG first builds a high-recall directed candidate graph from semantic, lexical, input/output, and structural evidence, with repair evidence and an optional LLM judge further refining candidate scores. It then applies direction-conditioned textual counterfactual probes that remove, substitute, and reorder skill pairs, aggregates the evidence with Bayesian smoothing, and publishes a state-filtered weighted graph for task-conditioned expansion. The graph is constructed offline and used without changing the downstream agent policy or task interface. Across six LLM backbones on ALFWorld ID-140 and ScienceWorld U211, CaSKG achieves the highest task score in all twelve combinations of model and benchmark. Relative to Graph-of-Skills (GoS), it improves the six-model macro-average ScienceWorld score from 72.62 to 80.50 and ALFWorld success from 80.01% to 86.79%, while reducing mean environment steps on both benchmarks. Qualitative and ablation analyses further show that calibrated edges help retrieval preserve prerequisites, state-changing actions, verification routines, and final completion steps. These results position edge-confidence calibration as an effective route to compact and executable skill retrieval at scale.</p>\n","updatedAt":"2026-08-28T02:55:56.694Z","author":{"_id":"6550bbc8c67f60a368a9a4b4","avatarUrl":"/avatars/dc66785028df83a83373de023e31d8dd.svg","fullname":"Zhiyuan Li","name":"zhiyuan218","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8747345805168152},"editors":["zhiyuan218"],"editorAvatarUrls":["/avatars/dc66785028df83a83373de023e31d8dd.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.25500","authors":[{"_id":"6a901e833bd48bb654ea6b42","user":{"_id":"6550bbc8c67f60a368a9a4b4","avatarUrl":"/avatars/dc66785028df83a83373de023e31d8dd.svg","isPro":false,"fullname":"Zhiyuan Li","user":"zhiyuan218","type":"user","name":"zhiyuan218"},"name":"Zhiyuan Li","status":"claimed_verified","statusLastChangedAt":"2026-08-27T16:45:04.784Z","hidden":false},{"_id":"6a901e833bd48bb654ea6b43","name":"Linyuan Gao","hidden":false},{"_id":"6a901e833bd48bb654ea6b44","name":"Xuechun Ding","hidden":false},{"_id":"6a901e833bd48bb654ea6b45","name":"Hongwei Chen","hidden":false},{"_id":"6a901e833bd48bb654ea6b46","name":"Yuan Wu","hidden":false},{"_id":"6a901e833bd48bb654ea6b47","name":"Yi Chang","hidden":false}],"publishedAt":"2026-08-26T00:00:00.000Z","submittedOnDailyAt":"2026-08-28T00:00:00.000Z","title":"CaSKG: Counterfactual-Causal Skill Graphs for Scalable Agent Skill Retrieval","submittedOnDailyBy":{"_id":"6550bbc8c67f60a368a9a4b4","avatarUrl":"/avatars/dc66785028df83a83373de023e31d8dd.svg","isPro":false,"fullname":"Zhiyuan Li","user":"zhiyuan218","type":"user","name":"zhiyuan218"},"summary":"Reusable skill libraries allow large language model (LLM) agents to reuse procedural knowledge across tasks, but they also turn memory access into a challenging retrieval problem. Full-library prompting preserves coverage at high context cost, vector retrieval returns compact neighborhoods but treats skills as independent text, and graph-based retrieval can recover workflow context only when the edges that carry relevance are reliable. We propose CaSKG, a counterfactual-causal skill graph framework that calibrates procedural relations before retrieval. CaSKG first builds a high-recall directed candidate graph from semantic, lexical, input/output, and structural evidence, with repair evidence and an optional LLM judge further refining candidate scores. It then applies direction-conditioned textual counterfactual probes that remove, substitute, and reorder skill pairs, aggregates the evidence with Bayesian smoothing, and publishes a state-filtered weighted graph for task-conditioned expansion. The graph is constructed offline and used without changing the downstream agent policy or task interface. Across six LLM backbones on ALFWorld ID-140 and ScienceWorld U211, CaSKG achieves the highest task score in all twelve combinations of model and benchmark. Relative to Graph-of-Skills (GoS), it improves the six-model macro-average ScienceWorld score from 72.62 to 80.50 and ALFWorld success from 80.01\\% to 86.79\\%, while reducing mean environment steps on both benchmarks. Qualitative and ablation analyses further show that calibrated edges help retrieval preserve prerequisites, state-changing actions, verification routines, and final completion steps. These results position edge-confidence calibration as an effective route to compact and executable skill retrieval at scaleCode is available at: https://github.com/ZhiyuanLi218/Caskg .","upvotes":2,"discussionId":"6a901e833bd48bb654ea6b48","githubRepo":"https://github.com/ZhiyuanLi218/Caskg","githubRepoAddedBy":"user","ai_summary":"CaSKG calibrates procedural skill relations via counterfactual-causal graph construction to improve compact, executable retrieval for LLM agents.","ai_keywords":["counterfactual-causal skill graph","directed candidate graph","textual counterfactual probes","Bayesian smoothing","state-filtered weighted graph","skill retrieval","LLM agents","procedural knowledge"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":0,"organization":{"_id":"67c1d682826160b28f778510","name":"antgroup","fullname":"Ant Group","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/662e1f9da266499277937d33/7VcPHdLSGlged3ixK1dys.jpeg"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6550bbc8c67f60a368a9a4b4","avatarUrl":"/avatars/dc66785028df83a83373de023e31d8dd.svg","isPro":false,"fullname":"Zhiyuan Li","user":"zhiyuan218","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"67c1d682826160b28f778510","name":"antgroup","fullname":"Ant Group","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/662e1f9da266499277937d33/7VcPHdLSGlged3ixK1dys.jpeg"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.25500.md","query":{}}">
CaSKG: Counterfactual-Causal Skill Graphs for Scalable Agent Skill Retrieval
Abstract
CaSKG calibrates procedural skill relations via counterfactual-causal graph construction to improve compact, executable retrieval for LLM agents.
Reusable skill libraries allow large language model (LLM) agents to reuse procedural knowledge across tasks, but they also turn memory access into a challenging retrieval problem. Full-library prompting preserves coverage at high context cost, vector retrieval returns compact neighborhoods but treats skills as independent text, and graph-based retrieval can recover workflow context only when the edges that carry relevance are reliable. We propose CaSKG, a counterfactual-causal skill graph framework that calibrates procedural relations before retrieval. CaSKG first builds a high-recall directed candidate graph from semantic, lexical, input/output, and structural evidence, with repair evidence and an optional LLM judge further refining candidate scores. It then applies direction-conditioned textual counterfactual probes that remove, substitute, and reorder skill pairs, aggregates the evidence with Bayesian smoothing, and publishes a state-filtered weighted graph for task-conditioned expansion. The graph is constructed offline and used without changing the downstream agent policy or task interface. Across six LLM backbones on ALFWorld ID-140 and ScienceWorld U211, CaSKG achieves the highest task score in all twelve combinations of model and benchmark. Relative to Graph-of-Skills (GoS), it improves the six-model macro-average ScienceWorld score from 72.62 to 80.50 and ALFWorld success from 80.01\% to 86.79\%, while reducing mean environment steps on both benchmarks. Qualitative and ablation analyses further show that calibrated edges help retrieval preserve prerequisites, state-changing actions, verification routines, and final completion steps. These results position edge-confidence calibration as an effective route to compact and executable skill retrieval at scaleCode is available at: https://github.com/ZhiyuanLi218/Caskg .
Community
Reusable skill libraries allow large language model (LLM) agents to reuse procedural knowledge across tasks, but they also turn memory access into a challenging retrieval problem. Full-library prompting preserves coverage at high context cost, vector retrieval returns compact neighborhoods but treats skills as independent text, and graph-based retrieval can recover workflow context only when the edges that carry relevance are reliable. We propose CaSKG, a counterfactual-causal skill graph framework that calibrates procedural relations before retrieval. CaSKG first builds a high-recall directed candidate graph from semantic, lexical, input/output, and structural evidence, with repair evidence and an optional LLM judge further refining candidate scores. It then applies direction-conditioned textual counterfactual probes that remove, substitute, and reorder skill pairs, aggregates the evidence with Bayesian smoothing, and publishes a state-filtered weighted graph for task-conditioned expansion. The graph is constructed offline and used without changing the downstream agent policy or task interface. Across six LLM backbones on ALFWorld ID-140 and ScienceWorld U211, CaSKG achieves the highest task score in all twelve combinations of model and benchmark. Relative to Graph-of-Skills (GoS), it improves the six-model macro-average ScienceWorld score from 72.62 to 80.50 and ALFWorld success from 80.01% to 86.79%, while reducing mean environment steps on both benchmarks. Qualitative and ablation analyses further show that calibrated edges help retrieval preserve prerequisites, state-changing actions, verification routines, and final completion steps. These results position edge-confidence calibration as an effective route to compact and executable skill retrieval at scale.
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2608.25500 in a model README.md to link it from this page.
Cite arxiv.org/abs/2608.25500 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2608.25500 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.