ConmuYan

kim-ccf-research-code

File-by-file coding with git-push workflow: implement, lint, commit, push. Outputs `code/` + `docs/dev_log.md`. Use for 编码, file-by-file coding, git push. Defer to kim-ccf-research-prototype for strict top-conference-style code from a baseline.

Testing & QA Listed

kim-ccf-research-experiment

Own experiment evidence semantics: decide datasets, baselines, metrics, ablations, robustness tests, chart evidence, and exactly what rows or columns a result table should contain. Use for experiment design, benchmark planning, supplied-result evidence structure, result-table schema, chart-spec semantics, 设计实验, 对比实验, 消融, 结果表证据结构. Do not search literature as the main deliverable, visually beautify or render an already specified table/figure, improve layout/colors/readability, or invent results.

kim-ccf-research-explore

Research direction exploration: from a fuzzy interest, converge to a research direction + RQ through multi-turn interaction. Outputs `docs/idea_report.md` Part 1. Use for 方向探索, 找方向, 调研, research direction. Do not write code, polish prose, or design experiments.

kim-ccf-research-humanize

Humanize and de-defend publication-facing CCF/AI prose and artifacts as a sidecar preflight, never as the primary task owner. Use before manuscript drafting/revision and before final manuscript experiment text, tables, captions, or method descriptions when Codex must remove defensive writing, boilerplate disclaimers, repetitive improbable edge cases, unnecessary safeguards, generic SHA-256/checksum requirements, duplicated smoke tests, simplified/toy methods, or AI-like risk narration. Do not auto-load as the sole owner for raw experiment planning, retrieval, review, auditing, routing, visual rendering, or assessment-only tasks without publication prose. Keep necessary concerns in a separate user-review warning instead of silently injecting them into files. Do not conceal material evidence, fabricate results, or override venue-mandated disclosures.

kim-ccf-research-idea-review

Strictly score, rank, compare, and triage early CCF research ideas with prior-art awareness and venue-fit risk. Use only for explicit idea scoring, score and rank multiple ideas, idea ranking, idea review, acceptance-potential triage, idea评分, 选题评分, 选题排名, 严格评审. Do not polish manuscripts, brainstorm directions, develop a fuzzy idea with no score, or optimize a single idea unless scoring is explicit.

kim-ccf-research-idea

Turn rough CCF research directions into concrete problem, gap, insight, method, novelty, and evidence plans. Use for idea optimization, fuzzy idea concretization, develop a fuzzy idea with no score, research direction shaping, early direction exploration, salvage routes, 优化idea, 具象化idea, 研究思路优化, 找方向, 方向探索. Do not rank or score multiple ideas as the main output.

kim-ccf-research-implement

Implementation design: produce a function-level implementation guide from the experiment plan and architecture sketch. Outputs `docs/implementation.md`. Use for 实现设计, function-level coding guide, module breakdown. Do not write the actual code; defer to kim-ccf-research-code.

kim-ccf-research-iterate

Code iteration loop: diagnose a bug or regression, backtrack to the responsible commit, update docs, modify code, re-test. Appends entries to `docs/dev_log.md`. Use for 代码迭代, debug, backtrack, diagnose. Do not restart from scratch; iterate the current version.

kim-ccf-research-literature

Search and screen external literature, related work, datasets, benchmarks, citation candidates, and research opportunity maps for CCF workflows. Use when external retrieval is the requested deliverable: literature search, related work discovery, prior art, benchmark search, direction scouting, 文献检索, 相关工作, benchmark搜索, 方向调研. Do not own supplied-result table schemas, experiment evidence structure, full manuscript review, already-cited-only audits, or manuscript writing.

kim-ccf-research-plan

Paper planning: update `docs/idea_report.md` part 4 (paper outline), choose figure plan (`notebooks/figures.ipynb`), select target venue (TPAMI/TKDE/AAAI/KDD/ICML/WWW/ICLR/CVPR), lock section ordering. Outputs the manuscript scaffold. Use for 论文规划, paper plan, outline. Defer to kim-ccf-research-write-* for section drafting.

kim-ccf-research-prototype

Implement or strip-down university-lab deep-learning paper prototypes in top-conference GitHub style: few files, readable forward/loss, easy to ablate and reproduce. Use when the user asks to implement a method, 实现算法, 写论文代码, 科研代码, 顶会开源风格, 最小实现, 去工程化, research prototype, or to remove factory/registry/hook/ trainer/audit wrappers. Also use for /research-code. Not for production services, web APIs, Docker/CI/packaging, or building a generic training framework.

kim-ccf-research-rebuttal

Write and organize CCF conference rebuttals, author responses, response letters, revision summaries, reviewer-comment ledgers, and conservative resubmission adaptation plans. Use for rebuttal, author response, AC/meta-review response, revision ledger, resubmission to a new venue, 审稿意见回复, 重投迁移. Do not trigger for ordinary manuscript writing.

kim-ccf-research-review

Own assessment-only manuscript deliverables without rewriting: full scientific review, scoring, reviewer reports, issue diagnosis, AC/meta-review, readiness judgment, writing/format review, and cross-version comparison. Use for full review, scientific review, do not rewrite, assessment-only, paper review, score drift, moving-target review, 完整审稿, 不要改写, 模拟审稿, 论文评分, 版本对比, 复审一致性, 写作评审, LaTeX检查. Requests for revised or polished prose, including rewrite based on reviews with no new review, belong to ccf-paper-writer; rebuttals belong to ccf-rebuttal-writer; visual/table styling belongs to ccf-visual-composer.

kim-ccf-research-shared

Shared infrastructure for the kim-ccf-research series. This is NOT a skill — it is a namespace providing cross-skill references (routing, task modes, handoff protocols, privacy controls, venue maps, artifact contracts). Every kim-ccf-research-* leaf loads files from ../kim-ccf-research-shared/references/. Never invoke this directly.

kim-ccf-research-submission

Check CCF conference submission readiness: venue LaTeX/template/page/anonymity/camera-ready rules, LaTeX/PDF build, metadata, fonts, supplementary files, artifact/reproducibility package, code/data/model release plan, licenses, seeds, and policy freshness. Use for submission check, venue format, NeurIPS/CVPR/ICML template/page limit, artifact package, 投稿检查, 会议格式. Do not polish manuscript content.

kim-ccf-research-visual

Own rendered visuals and visual redesign from supplied content or values: plot, beautify, lay out, generate, reconstruct, and QA paper figures, visual tables, method/architecture diagrams, icons, palettes, reference-guided layout control, and editable SVG/PDF/PPTX. Use for result-table layout, color/readability improvement, visual table redesign without changing numbers, 绘图美化, 排版, 配色, architecture diagrams, GPT Image 2 generation, reference-driven composition, explicit pure SVG, and editable reconstruction. Visual beautification remains here even for experiment results. Do not choose datasets/baselines/metrics, design evidence semantics, invent content, review the paper, rewrite prose, or convert PDFs into writing exemplars.

kim-ccf-research-write-abstract

Write the Abstract. Inputs: every other manuscript section once drafted. Outputs `docs/manuscripts/01_abstract.md`. Use for 撰写手稿 Abstract, 摘要. Should run LAST among write-* skills (after method, experiments, intro, related, conclusion are drafted).

kim-ccf-research-write-conclusion

Write the Conclusion. Inputs: every other manuscript section. Outputs `docs/manuscripts/05_conclusion.md`. Use for 撰写手稿 Conclusion, 结论. Should summarize contributions, limitations, and future work.

kim-ccf-research-write-experiments

Write the Experiments section. Inputs: idea_report, exp_plan, result tables/figures. Outputs `docs/manuscripts/03_experiments.md`. Use for 撰写手稿 Experiments, experiments section. Defers to kim-ccf-research-visual for figure rendering.

kim-ccf-research-write-introduction

Write the Introduction. Inputs: idea_report, problem framing, contributions list. Outputs `docs/manuscripts/00_introduction.md`. Use for 撰写手稿 Introduction, 引言. Standard CCF-A intro arc: motivate → gap → contribution → outline.

kim-ccf-research-write-method

Write the Method section. Inputs: idea_report, implementation.md, exp_plan. Outputs `docs/manuscripts/02_method.md`. Use for 撰写手稿 Method, method section. Defer to kim-ccf-research-write-experiments in parallel; defer to kim-ccf-research-humanize after draft committed.

kim-ccf-research-write-related

Write the Related Works section. Inputs: `docs/related_work_corpus.md` from kim-ccf-research-literature. Outputs `docs/manuscripts/04_related.md`. Use for 撰写手稿 Related Works, 相关工作. Must NOT run before kim-ccf-research-literature.

kim-ccf-research

Master orchestrator for the kim-ccf-research series. Detects the current research stage, routes to the correct leaf skill, and enforces shared cross-skill controls (handoff mode, task mode, evidence standards, privacy). Targets CCF-A venues (TPAMI, TKDE, AAAI, KDD, ICML, WWW, ICLR, CVPR) for ML/AI PhD work. Use when the user says /kim-ccf-research, asks where to start, names any kim-ccf-research-<purpose> leaf, or asks "what should I do next?" mid-paper. Trigger vocab: 研究助手入口, 论文流水线, research pilot, CCF-A 流水线, 论文路由, idea/literature /experiment/code/review pipeline, TPAMI/TKDE submission, 顶会投稿助手, 学术流水线入口, paper pipeline, research-to-publication.

kim-ccf-research-integrity

Audit CCF paper integrity: claim-support alignment, result-to-claim consistency, numeric consistency, terminology consistency, figure/table-to-text consistency, existing citation existence, BibTeX metadata, and citation-context support. Use for evidence audit, citation audit, consistency check, 引用核验, claim审计, 数字一致性. Do not perform full scientific review or broad literature search.

kim-ccf-research-translate

Chinese → English LaTeX translation of manuscript sections. Outputs `docs/manuscripts/<section>.en.md`. Use for 中英翻译, Chinese to English translation, 翻译. Run after every section is committed; do not interleave with kim-ccf-research-write-*.