{"id":19074,"plugin_id":"plugins_6a9174204a0481918ca3798d69d2e227","kind":"skill","collection_source":null,"comparison_source":null,"observed_at":"2026-09-30T23:15:14.456Z","digest":"34e8dc8b4a615256244f6f76c56041115e9c9c9dc3acaf49d19498c17f03e650","against":null,"payload":{"description":"Verify an external mathematical theorem, citation, notation translation, source-dependent implication, or bounded novelty claim, reusing authorized project-local source copies when available. Use when a proof relies on a named paper/result, when exact hypotheses or versions matter, when the user asks whether a claim is known, or when an authenticated mathematical source should be cached for later checks. Prefer primary sources and record the search scope. Do not treat snippets or failed searches as proof or global novelty.","included_files":[{"relative_path":"agents/openai.yaml","size_in_bytes":265},{"relative_path":"evals/evals.json","size_in_bytes":5141},{"relative_path":"evals/trigger-evals.json","size_in_bytes":705},{"relative_path":"references/source-cache.md","size_in_bytes":4564},{"relative_path":"references/source-record.md","size_in_bytes":998},{"relative_path":"scripts/literature_cache.py","size_in_bytes":31457},{"relative_path":"scripts/test_literature_cache.py","size_in_bytes":12955}],"name":"literature-check","skill_md_contents":"---\nname: literature-check\ndescription: >-\n  Verify an external mathematical theorem, citation, notation translation, source-dependent implication, or bounded novelty claim, reusing authorized project-local source copies when available. Use when a proof relies on a named paper/result, when exact hypotheses or versions matter, when the user asks whether a claim is known, or when an authenticated mathematical source should be cached for later checks. Prefer primary sources and record the search scope. Do not treat snippets or failed searches as proof or global novelty.\n---\n\n# Mathematical literature check\n\nVerify the exact implication, not merely the presence of related terminology.\n\n## Define the source question\n\nState:\n\n- the project claim or arrow requiring support;\n- likely source/result and acceptable source class;\n- required coefficients, grading, variance, finiteness, equivariance,\n  normalization, version, and range;\n- whether the task is theorem verification, attribution, notation translation,\n  overlap classification, or bounded novelty search.\n- for overlap or novelty, the terminology variants, older vocabulary, adjacent\n  fields, date horizon and citation graph likely to contain the same result.\n\nRead an existing literature-ledger entry and the dependent proof before\nsearching when they exist.\n\n## Acquire and authenticate\n\n1. When the project permits local source retention, query its literature cache\n   by exact DOI, arXiv version, ISBN, or other stable identifier before fetching.\n   A different or unversioned arXiv copy is only a discovery candidate.\n2. Prefer the published paper, official preprint, author manuscript, formal\n   documentation, or another primary source.\n3. Record title, authors, publication/preprint identifier, exact version or\n   revision date, stable locator, and date checked.\n4. Use abstracts, reviews, search snippets, lecture notes, and citation chains\n   only as discovery aids unless they are themselves the result being cited.\n5. For a changing preprint, verify that theorem numbering and hypotheses belong\n   to the version actually used by the project.\n6. Respect confidentiality and copyright; do not upload or reproduce licensed\n   or private material without authorization.\n\nAfter acquiring an authorized source, add it to the cache at a natural\ncheckpoint so later agents can reuse both the PDF and any extracted text. Read\n[source-cache.md](references/source-cache.md) before initializing or modifying\nthe cache. Cache hits save acquisition work; they do not authenticate the\nsource or verify its mathematical content.\n\n## Extract and translate\n\nRecord the exact theorem, definition, or formula used, including all hypotheses,\nexceptions, coefficient restrictions, source/target categories, variance,\nactions, grading, and completion assumptions. Note whether the source proves,\nsketches, states, conjectures, or only motivates it.\n\nWrite an explicit notation dictionary to the project conventions. Verify the\nproject implication one arrow at a time. A citation supplies no unstated\nfunctor, equivalence, coherence datum, normalization, or limiting argument.\n\nSeparate **source authentication**, **theorem extraction**, and **application\nto this project**. Record which of these was actually checked. A correctly\nidentified paper can still be inapplicable. Treat objectwise, natural,\nequivariant, filtered, integral and completed statements as different contracts\nuntil a comparison argument supplies the missing structure.\n\nUse [source-record.md](references/source-record.md) for durable entries.\n\n## Novelty and overlap\n\nRun discovery and verification as separate passes. In discovery, search the\nexact statement together with synonyms, older terminology, equivalent\nformulations and the names of the objects/invariants rather than only the\nproject's current title. Follow backward references from the closest source and\nforward citations when available; inspect relevant authors' earlier work and\nbibliographies in neighboring fields. Use more than one suitable index when\nfeasible and record which coverage was unavailable.\n\nIn verification, read the strongest candidates in their primary versions and\ncompare exact hypotheses and conclusion level. A title/abstract that appears\nadjacent can still contain the needed theorem, while matching terminology can\nhide an inapplicable result. For a material “apparently new” claim, use a second\nsearch strategy or fresh reviewer when available; disclose when the same searcher\nperformed both passes.\n\nClassify only as:\n\n- known verbatim;\n- known after translation of notation;\n- formal corollary not stated;\n- new proof of a known statement;\n- partial or adjacent result only;\n- apparently new within the stated search scope;\n- conjectural or explicitly open in a checked source.\n\nFor “apparently new,” report databases, exact and synonym queries, date range,\nlanguages or fields searched, backward/forward citation chains followed, the\nsecond-pass method, and important blind spots. A failed search is never a global\nnovelty theorem. Later-discovered overlap is a correction to append and propagate,\nnot a reason to rewrite the earlier scoped search as though it never occurred.\n\n## Record and report\n\nUpdate the project's literature ledger only when authorized and when the check\nchanges a dependency or attribution. Update status/log only if live research\nstate changes.\n\nWhen a `.mathbox/` ledger is in use, record a source evidence event through the\navailable `research-state` skill, pinning the durable extraction/translation\nreport. If a source version or interpretation changes, examine dependent claims\nand record a correction; do not overwrite the old check or silently refresh a\nhash. The cache's content hash alone is not a verified-source event.\nIf persistence is authorized but this host cannot execute or write, use the\n`research-state` deferred-handoff contract for a new extraction report and\nsource evidence proposal; state that local ingest has not recorded them.\n\nReport:\n\n1. exact source and version, plus the cache content hash and extraction\n   status when the source was retained locally;\n2. exact result used;\n3. notation/hypothesis translation;\n4. whether the implication is valid;\n5. overlap/novelty classification and search boundary;\n6. unresolved source ambiguity or missing implication.\n\nReturn **unverified** or **conditional** when the exact source is unavailable,\nthe translation fails, or the needed implication is neither stated nor formal.\n"},"changes":[],"summary":"First saved snapshot. No earlier version is available for comparison.","summary_kind":"deterministic","summary_metadata":{}}