{"id":41217,"plugin_id":"plugins_6a9174204a0481918ca3798d69d2e227","kind":"skill","collection_source":"plugin_package","comparison_source":null,"observed_at":"2026-10-24T12:58:02.384Z","digest":"f821962ff4f6abf5fe72919641d5b47655bd34596546e6c7ac84712f4faaa3f9","against":null,"payload":{"description":"Referee a whole mathematical manuscript for correctness, edge cases, notation, exposition and claim calibration. Use for a referee report, whole-paper stress test, or section-by-section manuscript audit. Do not use for a focused claim or proof audit, proofreading alone, a standalone citation or computation check, answering referee comments on one's own paper, or developing new proofs.","included_files":[{"relative_path":"agents/openai.yaml","size_in_bytes":275},{"relative_path":"evals/evals.json","size_in_bytes":15761},{"relative_path":"evals/trigger-evals.json","size_in_bytes":2633},{"relative_path":"references/final-referee.md","size_in_bytes":3293},{"relative_path":"references/math-scout-license.txt","size_in_bytes":1080},{"relative_path":"references/model-assignments.md","size_in_bytes":5676},{"relative_path":"references/output-contract.md","size_in_bytes":7411},{"relative_path":"references/preparation.md","size_in_bytes":3741},{"relative_path":"references/review-protocol.md","size_in_bytes":6614},{"relative_path":"references/reviewers/adversarial.md","size_in_bytes":1746},{"relative_path":"references/reviewers/claims.md","size_in_bytes":2082},{"relative_path":"references/reviewers/correctness.md","size_in_bytes":1897},{"relative_path":"references/reviewers/exposition.md","size_in_bytes":1470},{"relative_path":"references/reviewers/notation.md","size_in_bytes":1956},{"relative_path":"scripts/prepare_manuscript.py","size_in_bytes":33391},{"relative_path":"scripts/test_prepare_manuscript.py","size_in_bytes":31860}],"name":"referee","skill_md_contents":"---\nname: referee\ndescription: >-\n  Referee a whole mathematical manuscript for correctness, edge cases, notation, exposition and claim calibration. Use for a referee report, whole-paper stress test, or section-by-section manuscript audit. Do not use for a focused claim or proof audit, proofreading alone, a standalone citation or computation check, answering referee comments on one's own paper, or developing new proofs.\n---\n\n# Mathematical manuscript referee\n\nAssess the manuscript supplied using native reasoning and delegation; do not\ninstall external review runtimes, provider SDKs or credentials.\nDefault to leaving the manuscript unchanged. A review request authorizes review\nartifacts, not rewriting proofs, contacting authors or submitting a report.\n\n## Establish scope and prepare\n\n1. Read applicable project instructions and resolve the authoritative manuscript,\n   its version, source/project boundary, and requested review scope. Ask before\n   choosing between genuinely competing manuscripts or unclear boundaries.\n2. Focused obligations use the available specialists below; whole-paper coverage\n   and synthesis remain this skill's responsibility.\n3. Read [review-protocol.md](references/review-protocol.md) before reviewing.\n   Its severity, confidence, evidence and limited-context rules apply to every\n   pass, including synthesis.\n4. For LaTeX sources, use the optional standard-library helper following\n   [preparation.md](references/preparation.md). Resolve its installed directory;\n   manuscript paths are relative to the project, not this skill. For example:\n\n   ```bash\n   python3 <skill-directory>/scripts/prepare_manuscript.py paper.tex --root PROJECT --output RUN\n   ```\n\n   Use a new review directory in the project's designated review area, or\n   `referee/<run>/` at the project root if none is designated. Never write\n   review artifacts under `.mathbox/`: that directory belongs to a\n   research-state ledger, which pins evidence only from outside it. Creating a\n   review directory never initializes a ledger.\n5. Read the manifest, preparation limits and full source. Resolve preparation\n   errors without reading outside the authorized boundary; ask before widening\n   `--root`. Do not claim complete coverage until resolved. If only a PDF, pasted text or a\n   host without execution is available, build the same scope/coverage inventory\n   manually, use real page/result locators, and disclose extraction limits.\n\n## Survey the full manuscript\n\nIdentify central claims, hypotheses, conventions, internal dependencies,\nexternal-source leaves, and computational leaves. Map each main theorem to\nits load-bearing proofs and sections. Read abstract/introduction together with\nactual theorem statements. Extracted context is an index, not proof evidence;\nread referenced definitions and arguments in full.\n\nChoose review units by dependency structure: one section, a theorem and proof,\nseveral tightly coupled sections, or one whole-paper question. Include\nsubstantive front matter and appendices. Check units marked non-mathematical\nbefore skipping them; bibliography remains available to whole-paper passes.\n\n## Cover five dimensions\n\nRead each lane contract when assigning or performing that lane:\n\n| Lane | Scope and contract |\n|---|---|\n| Correctness | Every substantive unit and load-bearing proof; [correctness.md](references/reviewers/correctness.md) |\n| Adversarial | Concrete boundary cases, degeneracies, quantifiers, uniformity and reductions in those units; [adversarial.md](references/reviewers/adversarial.md) |\n| Exposition | Proof architecture, motivation and signposting for an expert; [exposition.md](references/reviewers/exposition.md) |\n| Notation | Whole manuscript, including definitions, conventions and cross-references; [notation.md](references/reviewers/notation.md) |\n| Claims | Whole manuscript, comparing headline promises with delivery and evidenced framing; [claims.md](references/reviewers/claims.md) |\n\nWhen user/project instructions prescribe models or reasoning, read\n[model-assignments.md](references/model-assignments.md); otherwise inherit host\ndefaults and record null requested fields. Use only exposed native controls;\ndisclose any fallback.\n\nFollow the host's delegation authorization rules. Skill invocation authorizes\nsubagents only if those rules permit it; where explicit user delegation is\nrequired, obtain it or use disclosed sequential self-review. Honor existing\nauthorization without asking again.\n\nFor each pass assign a unique `pass_id` and issue prefix, e.g. `corr-s3` and\n`corr-s3-F`. A child writes only `passes/<pass_id>.json`. Give it resolved paths\nfor [review-protocol.md](references/review-protocol.md), its lane file and snapshot\nunit files (paste text only if it cannot read files), exact source revision,\nnecessary dependencies and its disjoint scope. Withhold prior findings from\ncorrectness/adversarial passes; whole-paper lanes may receive unverified leads.\nChildren return exact specialist obligations; the coordinator arranges fresh\nconfirmation. They do not launch specialists unless explicitly assigned.\n\nSeveral compatible lanes may share a pass and coupled units may share an agent.\nBound concurrency by host capacity and mathematical usefulness; five-dimensional\ncoverage, rather than agent count, is the invariant. Apply model preferences\nthrough native controls, with fresh context when required. Allow context requests.\n\nWithout delegation, perform separate sequential passes and label their\nprovenance as self-review. Preserve raw returns and disclose failed, partial,\nunreviewed or reused scopes. Do not count a launch as completed coverage.\nRecord requested settings, observed execution settings and fallback reasons\nalongside coverage and structured findings using\n[output-contract.md](references/output-contract.md). An empty issue list is\nvalid. Keep low-confidence leads distinct from demonstrated defects.\n\n## Confirm focused obligations with Mathbox\n\nResolve available skills by name (`mathbox:<name>` in plugin installations,\nbare names when standalone). Read and follow the selected specialist contract.\nKeep this package independently installable: if a specialist is absent, perform\nthe bounded check directly with available tools, report that fallback, and\nleave anything not settled conditional.\n\n- **`proof-audit`:** use for serious mathematical findings that become concrete\n  obligations, such as whether Lemma 5.3 implies Proposition 5.7 under the\n  actual hypotheses. Supply the exact statements, dependency context and raw\n  argument for a fresh derivation. A lane flag alone is not confirmation.\n- **`literature-check`:** use when a conclusion depends on what an external\n  theorem establishes or on a material prior-art claim. State the bounded\n  source question, exact version/generalities needed and existing source\n  record. Follow its authorized cache-first workflow. Do not search every\n  citation. If the source is unavailable, retain a conditional obligation,\n  never a fabricated prior-art verdict.\n- **`computation-audit`:** use for load-bearing scripts, tables, enumerations\n  or numerical evidence. Compare claimed objects/population with actual\n  implementation and bounds. Run only justified authorized commands; bounded\n  success and sampled coverage do not prove a universal theorem.\n- **`proofread-math`:** use only for a requested mechanical proofreading\n  component. Keep it separate from exposition assessment and preserve the\n  review-only default unless manuscript edits were requested.\n- **`research-state`:** optional when an initialized ledger already exists\n  and provenance recording is useful and authorized. Pin actual statements,\n  evidence and reports from the review directory; ledger integrity does not\n  validate mathematics. Do not initialize it or create a parallel state\n  database merely for refereeing.\n\n## Reconcile and report\n\nFollow [final-referee.md](references/final-referee.md) and\n[output-contract.md](references/output-contract.md) for dispositions, concerns\nand the ordered report. Recheck raw findings against full source and specialist\nevidence; reviewer agreement is not proof. Keep raw returns separate from final\nconclusions, and disclose partial coverage. Absence of findings certifies neither\nwhole-paper correctness nor novelty.\n\nFor a revised manuscript, use the modest comparison/reuse rules in\n[preparation.md](references/preparation.md). Unchanged unit text is only a\ncandidate for reuse: check context, dependencies, lane-contract revision and\nactual prior artifacts. Rerun whole-paper passes and final synthesis on any\nsource change. Do not silently refresh old evidence or erase earlier findings.\n"},"changes":[],"summary":"First saved snapshot. No earlier version is available for comparison.","summary_kind":"deterministic","summary_metadata":{}}