{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:YQHAX3G6ZX6EBHIMC76ICVUKSC","short_pith_number":"pith:YQHAX3G6","schema_version":"1.0","canonical_sha256":"c40e0becdecdfc409d0c17fc81568a909cd036435d11b860ba0718823d90f827","source":{"kind":"arxiv","id":"2605.10913","version":2},"attestation_state":"computed","paper":{"title":"Shepherd: A Runtime Substrate Empowering Meta-Agents with a Formalized Execution Trace","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"Shepherd formalizes meta-agent operations as functions on a Git-like execution trace that records every interaction for fast forking and replay.","cross_cats":["cs.PL","cs.SE"],"primary_cat":"cs.AI","authors_text":"Ananjan Nandi, Christopher D Manning, Derek Chong, Dilara Soylu, Jiuding Sun, Simon Yu, Weiyan Shi","submitted_at":"2026-05-11T17:50:51Z","abstract_excerpt":"As LLM agent systems take on more complex tasks, they increasingly rely on meta-agents: higher-order agents that operate on other agents, much as managers supervise employees. Whatever a meta-agent does: coordinating agents, halting risky actions before execution, or repairing failed runs, requires manipulation of agentic execution at runtime. Existing agentic substrates make this hard: they give meta-agents only plain transcripts and environment snapshots, requiring it to build it's own tooling to reconstruct and orchestrate execution state. Therefore, we introduce Shepherd, a Python substrat"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":true},"canonical_record":{"source":{"id":"2605.10913","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.AI","submitted_at":"2026-05-11T17:50:51Z","cross_cats_sorted":["cs.PL","cs.SE"],"title_canon_sha256":"6fc3268d87e459f37de08123fcfdc086dd70475371d06456e3b5b0fa1b61f785","abstract_canon_sha256":"29bdb3a7d259864a973363d23fd598a091c5dc6647c42c2c9f168f44508d9368"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T02:05:10.346082Z","signature_b64":"/LDlmWa5jm0f+wnysPgGpLwAjqsF97+AOw/YMDqAzhKGCxBehql+8JTgfuXwl4fZyU55g5vdbkJkFwKPeK8CCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c40e0becdecdfc409d0c17fc81568a909cd036435d11b860ba0718823d90f827","last_reissued_at":"2026-05-26T02:05:10.345279Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T02:05:10.345279Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Shepherd: A Runtime Substrate Empowering Meta-Agents with a Formalized Execution Trace","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"Shepherd formalizes meta-agent operations as functions on a Git-like execution trace that records every interaction for fast forking and replay.","cross_cats":["cs.PL","cs.SE"],"primary_cat":"cs.AI","authors_text":"Ananjan Nandi, Christopher D Manning, Derek Chong, Dilara Soylu, Jiuding Sun, Simon Yu, Weiyan Shi","submitted_at":"2026-05-11T17:50:51Z","abstract_excerpt":"As LLM agent systems take on more complex tasks, they increasingly rely on meta-agents: higher-order agents that operate on other agents, much as managers supervise employees. Whatever a meta-agent does: coordinating agents, halting risky actions before execution, or repairing failed runs, requires manipulation of agentic execution at runtime. Existing agentic substrates make this hard: they give meta-agents only plain transcripts and environment snapshots, requiring it to build it's own tooling to reconstruct and orchestrate execution state. Therefore, we introduce Shepherd, a Python substrat"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"These results establish Shepherd as an efficient infrastructure for programming meta-agents.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The reported performance gains in runtime intervention, counterfactual optimization, and Tree-RL training are attributable to Shepherd's execution trace and forking features rather than unstated experimental factors.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Shepherd is a runtime system that formalizes meta-agent operations via typed execution traces, enabling fast forking and demonstrated improvements in agent intervention, optimization, and training on benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Shepherd formalizes meta-agent operations as functions on a Git-like execution trace that records every interaction for fast forking and replay.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"9e575e48cb623ae40487595aa4a1379367b17cfcb5a89a2b58cf4eacb860e7e4"},"source":{"id":"2605.10913","kind":"arxiv","version":2},"verdict":{"id":"8c35cbea-9b5f-4e8f-b37d-7132ab241a08","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-12T03:21:05.580474Z","strongest_claim":"These results establish Shepherd as an efficient infrastructure for programming meta-agents.","one_line_summary":"Shepherd is a runtime system that formalizes meta-agent operations via typed execution traces, enabling fast forking and demonstrated improvements in agent intervention, optimization, and training on benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The reported performance gains in runtime intervention, counterfactual optimization, and Tree-RL training are attributable to Shepherd's execution trace and forking features rather than unstated experimental factors.","pith_extraction_headline":"Shepherd formalizes meta-agent operations as functions on a Git-like execution trace that records every interaction for fast forking and replay."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.10913/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T05:02:00.955407Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T14:33:25.092429Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T10:31:17.245681Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T08:54:19.716782Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"ccbeafa3410c83c0718ee1f9118e0868f26df99e4c4c58cda2e976abd7d6eb10"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":1,"snapshot_sha256":"2c0819099842860587c1152b5a103b0c1bc2cc2e036588efef1b68988f1a7986"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2605.10913","created_at":"2026-05-26T02:05:10.345404+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.10913v2","created_at":"2026-05-26T02:05:10.345404+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.10913","created_at":"2026-05-26T02:05:10.345404+00:00"},{"alias_kind":"pith_short_12","alias_value":"YQHAX3G6ZX6E","created_at":"2026-05-26T02:05:10.345404+00:00"},{"alias_kind":"pith_short_16","alias_value":"YQHAX3G6ZX6EBHIM","created_at":"2026-05-26T02:05:10.345404+00:00"},{"alias_kind":"pith_short_8","alias_value":"YQHAX3G6","created_at":"2026-05-26T02:05:10.345404+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":1,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YQHAX3G6ZX6EBHIMC76ICVUKSC","json":"https://pith.science/pith/YQHAX3G6ZX6EBHIMC76ICVUKSC.json","graph_json":"https://pith.science/api/pith-number/YQHAX3G6ZX6EBHIMC76ICVUKSC/graph.json","events_json":"https://pith.science/api/pith-number/YQHAX3G6ZX6EBHIMC76ICVUKSC/events.json","paper":"https://pith.science/paper/YQHAX3G6"},"agent_actions":{"view_html":"https://pith.science/pith/YQHAX3G6ZX6EBHIMC76ICVUKSC","download_json":"https://pith.science/pith/YQHAX3G6ZX6EBHIMC76ICVUKSC.json","view_paper":"https://pith.science/paper/YQHAX3G6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.10913&json=true","fetch_graph":"https://pith.science/api/pith-number/YQHAX3G6ZX6EBHIMC76ICVUKSC/graph.json","fetch_events":"https://pith.science/api/pith-number/YQHAX3G6ZX6EBHIMC76ICVUKSC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YQHAX3G6ZX6EBHIMC76ICVUKSC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YQHAX3G6ZX6EBHIMC76ICVUKSC/action/storage_attestation","attest_author":"https://pith.science/pith/YQHAX3G6ZX6EBHIMC76ICVUKSC/action/author_attestation","sign_citation":"https://pith.science/pith/YQHAX3G6ZX6EBHIMC76ICVUKSC/action/citation_signature","submit_replication":"https://pith.science/pith/YQHAX3G6ZX6EBHIMC76ICVUKSC/action/replication_record"}},"created_at":"2026-05-26T02:05:10.345404+00:00","updated_at":"2026-05-26T02:05:10.345404+00:00"}