{"paper":{"title":"Skills as Verifiable Artifacts: A Trust Schema and a Biconditional Correctness Criterion for Human-in-the-Loop Agent Runtimes","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A skill is untrusted code until verified, so the runtime enforces verification before granting trust instead of relying on signatures or origins.","cross_cats":["cs.AI","cs.MA","cs.SE"],"primary_cat":"cs.CR","authors_text":"Alfredo Metere","submitted_at":"2026-05-01T05:53:05Z","abstract_excerpt":"Agent skills - structured packages of instructions, scripts, and references that augment a large language model (LLM) without modifying the model itself - have moved from convenience to first-class deployment artifact. The runtime that loads them inherits the same problem package managers and operating systems have always faced: a piece of content claims a behavior; the runtime must decide whether to believe it. We argue this paper's central thesis up front: a skill is untrusted code until it is verified, and the runtime that loads it must enforce that default rather than infer trust from a si"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"a skill is untrusted code until it is verified, and the runtime that loads it must enforce that default rather than infer trust from a signature, a clearance, or a registry of origin. Without skill verification, a human-in-the-loop (HITL) gate must fire on every irreversible call - which is operationally untenable and degrades into rubber-stamping at any non-trivial scale.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That a verification procedure exists which can satisfy the biconditional correctness criterion when evaluated on an adversarial-ensemble exercise, thereby allowing the capability gate to restrict HITL interventions to only unverified skills.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Proposes a trust schema including verification levels and a biconditional correctness criterion to verify skills in human-in-the-loop agent runtimes, reducing the need for constant oversight.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A skill is untrusted code until verified, so the runtime enforces verification before granting trust instead of relying on signatures or origins.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"7727d95b798b736ba4d8c5b86c9903cbfacd748eecd86f3d8b7f3c41d1e29947"},"source":{"id":"2605.00424","kind":"arxiv","version":2},"verdict":{"id":"4f22b992-87bd-441e-b81f-bc6c90dd0de3","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T18:20:20.231776Z","strongest_claim":"a skill is untrusted code until it is verified, and the runtime that loads it must enforce that default rather than infer trust from a signature, a clearance, or a registry of origin. Without skill verification, a human-in-the-loop (HITL) gate must fire on every irreversible call - which is operationally untenable and degrades into rubber-stamping at any non-trivial scale.","one_line_summary":"Proposes a trust schema including verification levels and a biconditional correctness criterion to verify skills in human-in-the-loop agent runtimes, reducing the need for constant oversight.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That a verification procedure exists which can satisfy the biconditional correctness criterion when evaluated on an adversarial-ensemble exercise, thereby allowing the capability gate to restrict HITL interventions to only unverified skills.","pith_extraction_headline":"A skill is untrusted code until verified, so the runtime enforces verification before granting trust instead of relying on signatures or origins."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.00424/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_compliance","ran_at":"2026-05-19T18:10:43.632512Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"26182b14ac943f843687b5b64549b3a7007200cd69721733e4ab8a7421018ccb"},"references":{"count":32,"sample":[{"doi":"","year":2025,"title":"Maksym Andriushchenko, Alexandra Souly, Mateusz Dziemian, Derek Duenas, Maxwell Lin, Justin Wang, Dan Hendrycks, Andy Zou, Zico Kolter, Matt Fredrikson, Eric Winsor, Jerome Wynne, Yarin Gal, and Xande","work_id":"6c6e61e1-3f05-48df-ba84-7c963b160bcd","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1976,"title":"Elliott Bell and Leonard J","work_id":"94d82569-6200-4beb-bd45-d57c94e071d1","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Alex Birsan. 2021. Dependency Confusion: How I Hacked Into Apple, Mi- crosoft and Dozens of Other Companies. https://medium.com/@alex.birsan/ 15 dependency-confusion-4a5d60fec610. Disclosure of a supp","work_id":"910ef2d1-e5c4-47ed-bd1f-56b804d3be21","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2008,"title":"Justin Cappos, Justin Samuel, Scott Baker, and John H. Hartman. 2008. A Look in the Mirror: Attacks on Package Managers. InProceedings of the 15th ACM Conference on Computer and Communications Securit","work_id":"592dea7a-ef42-42b6-8a55-d801b7dc50cd","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Zhaorun Chen, Zhen Xiang, Chaowei Xiao, Dawn Song, and Bo Li. 2024. AgentPoison: Red- Teaming LLM Agents via Poisoning Memory or Knowledge Bases. InAdvances in Neural Information Processing Systems (N","work_id":"72d623e5-a5f2-4892-ab54-561ca06e400b","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":32,"snapshot_sha256":"7c87d2148bfc794f0d15f06806e659ef11967ba84ba772b6f80fb8f122e99382","internal_anchors":6},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}