wanshuiyin

Quick single-paper lookup via AlphaXiv LLM-optimized summaries with tiered source fallback. Use when user says "explain this paper", "summarize paper", pastes an arXiv/AlphaXiv URL, or provides a bare arXiv ID for quick understanding - not for broad literature search.

Search, download, and summarize academic papers from arXiv. Use when user says "search arxiv", "download paper", "fetch arxiv", "arxiv search", "get paper pdf", or wants to find and save papers from arXiv to the local paper library.

Autonomously improve a generated paper via GPT-5.6-Sol xhigh review → implement fixes → recompile, for 2 rounds. Use when user says "改论文", "improve paper", "论文润色循环", "auto improve", or wants to iteratively polish a generated paper.

Search and progressively read open-access academic papers through DeepXiv. Use when the user wants layered paper access, section-level reading, trending papers, or DeepXiv-backed literature retrieval.

AI-powered web search via Exa with content extraction. Use when user says "exa search", "web search with content", "find similar pages", or needs broad web results beyond academic databases (arXiv, Semantic Scholar).

SSH job queue for multi-seed/multi-config ML experiments with OOM-aware retry, stale-screen cleanup, and wave-transition race prevention. Use when user says "batch experiments", "队列实验", "run grid", "multi-seed sweep", "auto-chain experiments", or when /run-experiment is insufficient for 10+ jobs that need orchestration.

Generate deterministic publication-quality architecture, workflow, and pipeline diagrams from structured JSON (FigureSpec) into editable SVG. Use when user says "架构图", "workflow 图", "pipeline 图", "确定性矢量图", "figure spec", "draw architecture", or needs precise, editable, publication-ready vector diagrams. Preferred over AI illustration for formal architecture/workflow figures.

Draft a structured grant proposal from research ideas and literature. Supports KAKENHI (Japan), NSF (US), NSFC (China, including 面上/青年/优青/杰青/海外优青/重点), ERC (EU), DFG (Germany), SNSF (Switzerland), ARC (Australia), NWO (Netherlands), and generic formats. Use when user says "write grant", "grant proposal", "申請書", "write KAKENHI", "科研費", "基金申请", "写基金", "NSF proposal", or wants to turn research ideas into a funding application.

Phase 2/3 of a movie — bake + cross-model-verify a movie from an authored comic.json. Per frame: render the content-SVG blueprint, bake via the agent mcp__codex__codex sidecar, gate with a 3-reviewer cross-model panel (narrative [currently the codex CLI] ‖ Gemini visual ‖ Codex visual, deterministic fuse), keep/retry/assemble, write the wiki trace, project to the viewer. The movie-side twin of method-figure's render half. (Author the comic.json first with the comic-author skill.)

Generate a publication-grade method / architecture / pipeline / workflow figure (a paper or README 'Figure 1') as an AUDITABLE object, not a one-shot prompt. A deterministic JSON blueprint LOCKS the content; an image model (gpt-image-2, baked by the agent via mcp__codex__codex — Codex GPT-5.5 xhigh, sandbox workspace-write) bakes the aesthetic from a labeled-condition render + the project's real identity refs; a cross-model panel (Gemini + Codex) blind-transcribes the result and a script hard-diffs it against the blueprint; the loop regenerates until Gemini AND Codex approve and the diff is empty — then the calling agent (Claude) gives the structural sign-off. NOT for statistical plots (use a plotting tool) or photo scenes.

Phase-1 (S4) of a comic movie — PRODUCE the canonical reusable references the whole spiral conditions on. Per asset it bakes ONE canonical 1:1 white-bg identity ref via the agent mcp__codex__codex sidecar bake (Codex native image_gen — conditioned, never hand-pasted) OR, for a deterministic motif (clock/chart/stamp/star-map), emits a single-source parametric SVG from a python generator (asset_lib.py). Hashes + base64-encodes the bake into the asset node's output_ref (all 6 fields or it's a schema violation), versions every ref _v{NNN} with a supersedes self-edge, and runs a single-source collision gate. GENERATES ONLY — it NEVER self-locks/approves (that is the cross-model asset-review-loop's job, a different model family). Identity is BRING-YOUR-OWN; the ARIS chibi duo is only the worked example.

Phase-1 (S5) UPSTREAM ref-asset gate — the bounded cross-model adversarial loop that LOCKS one reusable identity-locked asset (character sheet / location plate / prop cutout / text-panel / logo-free symbol) BEFORE it can be composited into any panel. This is NOT the panel_gate (composed-output) or the assembly_gate (cross-panel) — it gates the refs that FEED both, so identity drift is caught at the source, not downstream. Two layers: (1) a MANDATORY static single-source collision check (check_asset_collisions.py — 'one visual dialect, never two', enforced by tool) + zero-text/literal build-asserts; (2) a per-round 3-reviewer panel (Claude narrative ‖ Gemini visual ‖ Codex synth) that blind-scores 5 dims and LOCKS only at 准×3 unanimity (prior_lock_count>=3 approvals in the SAME round) — else regenerate (route to comic-asset-ref-generator) or, at MAX_REVIEW_ROUNDS, escalate the asset REQUIREMENT back to outline/storyboard. Hard IP veto. Use when you say 'lock this character ref', 'asset gate', 'ref review', 'is

Phase 1 ORCHESTRATOR of a movie/comic — turn a fuzzy story idea into the Authored Source of Truth (a schema-valid comic.json + its locked asset library) by driving the detailed author skills in order (intent → style → outline → storyboard → assets → blueprints → prompts → comic.json), each gated by comic-cross-layer-gate, so comic-director (Phase 2/3) can bake + cross-model-verify it. You don't hand-write comic.json; this is the workflow your agent runs to author it. Use when the user says "做个漫画/电影", "from this idea make a comic", "author the comic.json", "run the comic pipeline".

Phase-1 comic-author step (post-final eval) — a DOUBLE-BLIND A/B of two FINAL whole comics: our cross-model-audited progressive render (the comic-author + comic-director output) vs a naive single-shot baseline. A single sealed coin-flip hides which is which; two cross-model reviewers (Codex + Gemini) score both on a fixed rubric reading only a SHARED blind spec (intent + ART_BIBLE); only AFTER both reviews land do we unseal, re-label, and write a Chinese comparison.md + the A/B verdict nodes. editability/traceability is the structural wedge that can win even when the baseline looks prettier. Use when the user says "和 baseline 比", "blind comparison", "A/B 评测", "对比 baseline", "盲评", "whole-comic vs one-shot", or a finished comic needs a baseline-relative honest verdict. NOT an authoring skill and NOT the per-panel panel_gate / assembly_gate — those run DURING production; this runs AFTER, on two complete works.

Phase-1 (S7 of the comic-author suite) — turn ONE locked panel_spec into a deterministic content-SVG blueprint that becomes the bake condition (reference #1 of the agent mcp__codex__codex sidecar bake), by WRITING A PYTHON GENERATOR (never raw SVG in chat — LLMs botch coordinates). The HEADLINE comic-pivot rule: the image bakes NO bubbles at all — draw only characters + scene + leave negative-space SAFE ZONES; HTML/CSS owns the entire bubble. Every panel gets a content_svg (a figure OR a layout blueprint); a baked figure-panel must declare expected_literals verbatim. Single-source collision check is mandatory. Gated by comic-cross-layer-gate --gate blueprint. Use when the storyboard is locked and you need each panel's deterministic generation condition; do NOT use to bake the panel (that is comic-director) or to write the verdict-stamp/curve assets (that is comic-asset-ref-generator).

Phase-1 (S9, the FINAL comic-author step) — assemble the LOCKED storyboard + locked per-panel blueprints into ONE schema-valid `comic.json` (the comic-ir/1.0 contract boundary handed to comic-director / run_comic.py). Project the page_order + each panel's condition{} + render fields into `pages[]` + keyed `panels{}` per schemas/comic.schema.json; author ONLY authored fields and leave image_path/active_attempt_id/wiki_node_id EMPTY for the engine. THE step where page-count integrity is reconciled and the orphan-panel class of bug is caught (a panel defined in panels{} that no page references — ship that and the finale silently vanishes) — caught by an INLINE whole-comic reconcile this skill runs (page_refs vs panels{} keys vs the storyboard page_order), because the deterministic scripts are per-page and the schema leaves condition/content_svg OPTIONAL. Gated by comic-cross-layer-gate --gate compile (the DETERMINISTIC gate: both run_comic.py --dry-run AND cli/validate_wiki.py must exit 0). Use when the user say

End-to-end substantive-integrity forensic sweep of a research paper (especially autoresearch / AI-Scientist-style output). Orchestrates the whole pipeline: ingest (arxiv-id | pdf | dir → working dir + pdftotext for L0) → /evidence-ledger (artifact manifest + observability level L0/L1/L2 + span-anchored claims.json) → fan out the integrity auditor skills (consistency, citation, baseline, experiment, presentation, proof-derivation, eval-design — each reads the ledger, emits span-anchored findings) + the zero-verdict-weight AIS writing-style track → advisory memos (/adversarial-case-builder + /novelty-duplication-advisory, no verdict weight) → deterministic tools/adjudicate_findings.py (--ledger REQUIRED) → reviewer-ready Integrity Forensics Report. Cross-model (fresh codex per dimension) and reviewer≠adjudicator: the model proposes findings, the deterministic adjudicator decides the verdict. Observability-aware, detect-only, never an opaque AI-text classifier (a separate zero-weight AIS section lists AI writing

Flagship intra-paper self-consistency forensics: does the paper contradict ITSELF across abstract/intro/tables/body/appendix, and does the method DESCRIBED match the method EVALUATED? Needs no external ground truth — works PDF-only (L0). Runs a deterministic arithmetic pass + a fresh cross-model semantic pass, every finding span-anchored to the evidence ledger (claims.json), reviewer≠adjudicator. Emits consistency-audit.findings.json; NEVER computes the verdict. Triggers: "consistency audit", "check the paper against itself", "self-consistency", "内部自洽".

Audit whether a paper's baseline comparisons are COMPLETE, FAIR, and SIGNIFICANT: a required recent SOTA baseline is missing while 'best/SOTA' is claimed (HP-MISSING-BASELINE); a baseline is undertuned / given less compute-tuning-data, run at a mismatched config, or the equal-budget ablation-as-baseline is absent (HP-WEAK-BASELINE); 'outperforms' is asserted over overlapping error bars or with no variance/seeds (HP-SIG-OVERLAP); and a cross-row 'improves over baseline by X%' is arithmetically wrong (HP-DELTA-ERROR, cross-row form only). A versioned per-domain baseline profile + a live leaderboard/recency search are assembled by the EXECUTOR as structured facts; a fresh cross-model reviewer (gpt-5.6-sol xhigh, read-only, fresh thread per dimension) PROPOSES findings, each span-anchored to a ledger claim_id; tools/adjudicate_findings.py DECIDES the verdict. Works at L0 (stated comparisons) and deepens at L2 (configs/result files). A completeness question it cannot settle internally becomes needs_external_check,

Citation-integrity forensics: is every reference real, correctly attributed, and used in a context the cited work actually supports? Catches hallucinated references (no paper at the claimed arXiv id/DOI/venue, fabricated authors/year), metadata drift (wrong year/venue/version), and wrong-context citations (a real paper cited for a claim it never makes — or argues against). A hot zone for machine-generated papers. Decidable at L0 (text + canonical sources). Span-anchored to the evidence ledger (claims.json); the executor gathers canonical facts (DBLP / arXiv / DOI), then one FRESH cross-model thread per cited key proposes findings; reviewer != adjudicator. Emits citation-forensics.findings.json; NEVER computes the verdict. Triggers: "citation forensics", "check the references", "hallucinated citations", "wrong-context citation", "verify references", "引用核对".

Audit whether a paper's EVALUATION DESIGN actually measures what it claims and whether its reporting is complete — the validity layer family D (experiment-forensics) cannot reach. Three patterns: train/test leakage means the reported score may not measure generalization (HP-EVAL-LEAKAGE — adopts the Kapoor & Narayanan 8-type / 3-category leakage taxonomy; the illegitimate-proxy / sampling-bias / pretraining-contamination subtypes hand off as needs_external_check, naming but NEVER running Oren-2023 exchangeability / Shi-2023 Min-K% / Golchin-2023 Time-Travel / BIG-bench canary); a load-bearing LLM judge is conflicted (same model/family as a compared system) or unvalidated (no human-agreement, no bias control) (HP-JUDGE-VALIDITY); a declared condition/metric is dropped or switched to favor the method, or 'best' is chosen with no held-out set (HP-SELECTIVE-REPORTING). Verdict-bearing at L0/L1 from the DESCRIBED protocol — NOT repo-gated like experiment-forensics; L2 only CONFIRMS against split/preprocessing/resu

Audit experiment integrity against the evidence ledger. At L2 (repo + result files present) a fresh cross-model reviewer reads the eval code line-by-line for fake/derived ground truth, score self-normalization, phantom results (a paper number with no backing file/key), dead/uncalled metric code, verified-scope inflation, method-described ≠ method-evaluated drift, synthesized-looking results, placeholder/fake data still wired into a released result, code-output ≠ reported-number mismatch, and missing reproducibility artifacts (an empirical/agent/LLM paper shipping neither code nor the prompts/configs its results need) — every finding span-anchored to a ledger claim_id. At L0/L1 (PDF / source only) the same patterns are surfaced as info-level 'could-not-verify' signals where the ledger gives an anchor (observability_level_required:2) — NEVER a fraud verdict from a PDF. The reviewer PROPOSES findings; tools/adjudicate_findings.py computes the verdict. Detect-only. Triggers: "experiment forensics", "audit the res

Family-G proof & derivation integrity forensics: does a THIRD PARTY's written proof/derivation actually establish its theorem, or does it skip an obligation, assume its own conclusion, take an invalid step, drift a symbol's meaning, or smuggle an unstated assumption? Decides from the WRITTEN proof/derivation — verdict-bearing at L1 (the LaTeX source; PDF-extracted math is unreliable, so an L0 PDF-only run surfaces info only) — never asserts 'fabricated', only that the step shown does not hold. A fresh cross-model reviewer reads the theorem/proof + an extraction-only obligation scaffold and proposes per-obligation findings, each span-anchored to the evidence ledger (claims.json); reviewer≠adjudicator. Emits proof-derivation-forensics.findings.json; NEVER computes the verdict. dimension=proof, can be critical. Triggers: "proof forensics", "check this proof", "derivation integrity", "audit the math", "证明审计", "推导有没有漏洞".

Synthesize the single strongest EVIDENCE-BOUND reviewer case to reject a paper, built ONLY from the evidence ledger (claims.json) + the other auditors' confirmed findings — never free-floating LLM critique. Two fresh cross-model codex threads: an attack writes the ~200-word rejection paragraph (every accusation tagged to an existing claim_id/finding_id), a defense decomposes it and rules each point against the anchored evidence. MEMO-ONLY: emits adversarial-case-builder.memo.md (fed to the adjudicator via --memo) and carries NO verdict weight — tools/adjudicate_findings.py lists it in MEMO_ONLY_SKILLS and caps it at info. Honest-null allowed (the paper may survive). Run LAST. Detect-only. Adapted from ARIS kill-argument. Triggers: "adversarial case", "strongest objection", "rejection memo", "kill argument", "最强拒稿点".

Render an ARIS Markdown / JSON artifact (IDEA_REPORT, AUTO_REVIEW, KILL_ARGUMENT, PAPER_PLAN, research-wiki state, etc.) into a single-file HTML view designed for human reading. Academic template outputs are gated by a fresh cross-model Codex review for render fidelity + safety (the ARIS invariant). Use when the user says "渲染 HTML", "出一份 HTML 报告", "render html", "make this readable", "export to html", or wants a polished web-rendered view of a Markdown artifact. Markdown/JSON stays the canonical source; HTML is a generated, reviewed view.

Generate a long-form Chinese interview-prep cheat sheet on a specific ML/LLM topic — formulas with derivations, from-scratch PyTorch code, comparison tables, and 25 高频面试题 (L1 必会 / L2 进阶 / L3 顶级 lab). Cross-model codex review checks math, code, historical citations, and style discipline; then /render-html produces a single-file HTML with academic-newspaper template. Output: docs/tutorials/<slug>_tutorial.{md,html,review.json}. Use when the user says '写面试 cheat sheet', '写一份 X 教程', '帮我准备 Y 面试题', '出一份 X 速查', or wants a 600-1000 line Chinese tutorial on a specific ML topic.

Generate a fact-checked academic personal homepage from a CV, optionally augmented by an existing manual homepage and an assets directory. Produces editable structured source files (profile.yml + publications.bib + bio.md + news.md) and a single-file HTML page. Uses Codex MCP for independent factual review against DBLP / arXiv. Optionally uses Gemini multimodal for screenshot critique when available. Use when the user says '做个学术主页', '从CV生成主页', 'aris-homepage', 'generate academic homepage from CV', 'PhD homepage', 'GitHub Pages personal site', or wants a fact-checked academic site.