Research

Diagnose surprising or negative ML/AI results. Use when methods fail, metrics conflict, seeds vary, baselines win, plots look suspicious, or next action is unclear.

Use when auditing completed results for confounds, claim-drift, protocol integrity, or attribution before locking claims into the paper. Not for deciding what to do after a surprising result (use result-diagnosis). Not for significance tests or effect sizes (use statistical-analysis-planner). Not for engineering failures (use experiment-debugger).

Initialize, inspect, and maintain a hierarchical memory system for an ML research project across paper, code, worktrees, slides, reviewer simulation, rebuttal, experiments, claims, evidence, risks, and actions. Use this skill whenever the user wants cross-session project memory, project bootstrapping context, feedback-loop tracking, claim-evidence-risk-action alignment, worktree memory, or consistency between code results, paper writing, slides, reviews, and rebuttal.

Help a CS or AI PhD student turn a rough research idea into a validated next-step decision using the FIVE+C framework. Use this skill whenever the user says they have a research idea, wants to know whether an idea is worth pursuing, needs help choosing between project directions, is preparing to pitch an idea to an advisor or senior student, or feels unsure whether a project is too incremental, too ambitious, already solved, hard to evaluate, or missing resources.

Coordinate local, Git remote, and SSH/HPC/RunAI research projects. Use for server state, sync safety, job submission, interactive sessions, logs, artifact lookup, context recovery, raw SSH commands, remote shell one-liners, SSH quoting issues, remote-cmd, remote-bash, or avoiding local shell expansion of remote variables.

Draft ML/AI related work as novelty-boundary writing. Use for closest-work grouping, citation roles, paragraph plans, boundary statements, and safe novelty wording.

Read and summarize project reference sources into structured source cards. Use for skimming papers, PDFs, Word docs, Markdown notes, BibTeX files, scripts, specs, collaborator feedback, or source bundles; extract writing patterns, methods, theory, benchmarks, baselines, implementation hints, risks, constraints, and project seeds without yet deciding project implications.

Connect structured reference source cards to the active ML project. Use when papers, collaborator docs, Markdown notes, specs, scripts, BibTeX files, or source bundles should inform claims, risks, baselines, benchmarks, experiments, algorithm design, implementation, writing contracts, citations, collaborator actions, project initialization, or memory writeback.

Manage project reference sources under reference/. Use when scanning, ingesting, indexing, deduplicating, monitoring, or tracking processing status for papers, PDFs, Word docs, Markdown notes, BibTeX files, scripts, specs, or source bundles without deeply reading them.

Produce a multi-paper comparison matrix across a literature corpus with tiered read depth. Use when multiple papers need to be compared side-by-side for method differences, performance gaps, closest-work ranking, or trend identification — distinct from per-paper source cards (reference-reading-summarizer) and single-paper project linking (reference-project-synthesizer).

Plan and write strategic rebuttals after real paper reviews arrive. Use this skill whenever the user has OpenReview reviews, reviewer comments, scores, confidence ratings, meta-reviews, author response windows, or wants to decide which experiments to run, infer reviewer intent, draft point-by-point responses, prepare follow-up discussion replies, or improve wording after reviews for ML/AI venues such as NeurIPS, ICML, ICLR, CVPR, ACL, EMNLP, or similar conferences.

Sync experiment results from the code repo into the paper's daily experiments log (daily_experiments.tex). Use when you have new experiment results to record, want to update the paper with latest numbers, or log experimental findings from an ML research project.

Plan mid-project direction changes when consistent negative results or novelty challenges require scope narrowing, angle change, or kill decisions. Use after multiple result-diagnosis cycles fail to recover the original claim. Distinct from research-idea-validator (project start) and result-diagnosis (per-experiment).

Initialize a new ML research project with aligned paper (LaTeX) and code (Python) repositories under a shared parent folder. Use when starting a new research project, setting up a paper+code workspace, or initializing a new ML research environment.

Maintain automatic personalization writeback from agent trajectories, logs, sidecar artifacts, and repeated user preferences. Use when a task produces reusable preferences, lessons, private user memory, project contracts, or candidate public skill rules without interrupting the user.

Use to track nonlinear drafting state — section status, claim-text dependencies, stale prose, style decisions, and edit impact across sessions. Not for writing prose (use paper-writing-assistant). Not for planning the initial writing contract (use paper-writing-contract-planner).

Create a paper writing contract before drafting. Use to lock venue, positioning, archetype, section order, paragraph roles, evidence slots, figure/table jobs, and forbidden claims.

Draft and revise ML/AI paper prose as a claim-aware writing assistant. Use for section writing, result interpretation, venue-aware style, and provisional metrics.

Simulate target-conference reviewers for an ML/AI paper before submission. Use this skill whenever the user wants a reviewer-style critique, predicted scores, likely reject reasons, rebuttal risks, area-chair style meta-review, adversarial Reviewer 2 feedback, or venue-specific pre-review for conferences such as NeurIPS, ICML, ICLR, CVPR, ACL, EMNLP, or similar venues. This skill should dynamically inspect reviewer guidelines, example reviews, accepted papers, and project evidence when availa...

Decide what an ML or AI paper should strategically sell before detailed writing or venue-specific polishing. Use this skill whenever the user has an idea, literature map, experiment results, figures, reviewer risks, or a draft and needs to choose the paper's primary contribution, claim scope, paper archetype, target audience, novelty framing, related-work boundary, title/abstract/main-figure story, or claims to avoid before using conference-writing-adapter.

Plan and draft ML/AI introductions as venue-aware argument chains. Use for hook, gap, insight, method, result, contribution flow, and paragraph roles.

Maintain a paper-facing evidence board that aligns claims, experiments, figures, tables, sections, reviewer risks, and next actions during ML/AI paper writing. Use this skill whenever writing exposes missing experiments, new results require paper changes, reviewer simulation reveals evidence gaps, claims need support checks, figures/tables need mapping to claims, or the user wants a live paper evidence board before submission.

Edit ML/AI paper drafts for internal consistency. Use after sections exist to align claims, terminology, figures, tables, captions, limitations, and conclusion.

Audit private skills, memories, notes, or operational logs before turning them into public skills, templates, docs, or reusable patterns. Use when scanning personal/private memory for publishable knowledge, redaction needs, privacy risks, source-visibility leaks, or PR-ready public skill candidates.