All authors
Claude Skills by yulonglin
github.com/yulonglin61 skills7 installs203 views
- DeslopStrip AI-generated slop and overengineering out of code you or an agent just wrote — narrating comments a human would never add, defensive try/except and null guards on already-validated codepaths, `any` casts papering over type errors, needless abstraction layers, and any style the rest of the file does not use. Use before committing agent-written code, when reviewing an uncommitted diff, and on "deslop", "remove the slop", "clean up this diff", "de-AI this code", "it looks LLM-written", "st...Votes: 0GitHub stars: 13
- Mats SlurmSLURM helpers for the MATS compute cluster — interactive and batch GPU jobs on A100 and L40, queue and GPU-availability monitoring, log tailing, and the one-time alias setup. Use when submitting or watching a job on MATS, when `grun`, `gbatch`, `arun`, `jlog` or `te` appear, or when a cluster job is queued, pending or failing to get a GPU.Votes: 0GitHub stars: 13
- Read PaperRead a research paper or technical post with critical rigor — what the authors actually claim, which methodological choices drive the result, how strong the evidence is, and where the paper sits against related work. Use when reading a PDF or arXiv link, doing a literature review, evaluating a method or a claim, comparing approaches across papers, when the user shares a paper URL or file, or when explaining a concept that requires reading the source ("explain GRPO", "what is constitutional AI...Votes: 0GitHub stars: 13
- Spec Interview ResearchInterview Yulong into a research spec before an experiment is built — hypotheses and what would falsify them, independent and dependent variables, controls and confounds, baselines, datasets, sample size and power, resources, reproducibility. Use when starting a new experiment, an ablation, an A/B test or a benchmark run, when a design needs its assumptions challenged before any code is written, or on "spec this experiment", "help me design this run", "interview me about this experiment".Votes: 0GitHub stars: 13
- Tikz DiagramsDraw or revise a conceptual diagram for a LaTeX paper — pipelines, architectures, eval flows, training loops, comparison panels — in Anthropic's pastel minimalist style. Use when a paper or preprint needs a figure that is not a data chart, when adapting a blog-post diagram for NeurIPS/ICML/ICLR, or when an existing TikZ figure needs restyling. Carries 17 pattern skeletons, a semantic colour scheme, the anthropic-tikz style package and 16 reference images to aim at.Votes: 0GitHub stars: 13
- Externalise HandoverExternalise the important parts of this conversation into a handover document for another person or agent — next tasks, what was accomplished with exact commands and output paths, bugs hit, open uncertainties, and the user's own instructions cleaned up. Use on "hand this over", "write a handover", "handoff brief", "brief another agent on this", "externalise this conversation", or /externalise-handover.Votes: 0GitHub stars: 13
- Clear WritingThe clarity standard for papers, reports, artifacts, specs and updates — paragraph structure, sentence mechanics, claim calibration, and the LLM tics to cut. Use when drafting or revising anything a reader must follow.Votes: 0GitHub stars: 13
- Research PresentationThe standard for how research is presented — structuring a research presentation, slide deck, research update, mentor or team meeting, a report or an artifact page. Use when building, structuring or critiquing slides, a research update, a findings page, or a figure destined for a deck.Votes: 0GitHub stars: 13
- Review DraftRun the critic panel over a draft and merge what comes back — clarity, narrative, facts, red-team, in parallel, each writing a feedback file. Use on "review my draft", "critique this post", "give me feedback on this write-up", "run the critics", "tear this apart", or before publishing a blog post, LessWrong post, explainer, report or announcement. Also use when only one lens is wanted (just clarity, just fact-checking), since this skill dispatches the individual critics too.Votes: 0GitHub stars: 13
- Review PaperThis skill should be used when the user asks to "review a paper", "critique a manuscript", "give feedback on a draft", "review this write-up", "analyze paper quality", or wants constructive feedback on ML/AI research writing based on Neel Nanda's paper-writing criteria.Votes: 0GitHub stars: 13
- SlidevSlidev deck tooling — build and preview a deck, export it to PDF or PNG, and fix a slide whose content overflows, is cut off, or renders as a blank page. Use when creating or editing slides, when the user mentions Slidev, slides.md, `pages/*.md` slide files, a dev server for slides, or a PDF/PNG export; and equally when a slide overflows, text runs off the edge, a page comes out blank, or the user asks to fix or clean up slides in an existing deck. Slidev-only — it drives `bunx slidev`.Votes: 0GitHub stars: 13
- Strategic CommunicationGuide for effective communication in situations requiring negotiation, persuasion, or strategic thinking. Built on Chris Voss's FBI negotiation approach - tactical empathy and emotion labeling as primary tools. Use when drafting messages for rentals, salary discussions, work requests, declining offers, or any situation where you need to be both warm and strategically clear.Votes: 0GitHub stars: 13
- My InsightsRun custom usage analytics on all Claude Code sessionsVotes: 0GitHub stars: 13
- Skill CreatorGuide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Codex's capabilities with specialized knowledge, workflows, or tool integrations.Votes: 0GitHub stars: 13
- Skill InstallerInstall Codex skills into $CODEX_HOME/skills from a curated list or a GitHub repo path. Use when a user asks to list installable skills, install a curated skill, or install a skill from another repo (including private repos).Votes: 0GitHub stars: 13
- Commit Push SyncThis skill should be used when the user asks to "commit and push", "commit push", "sync changes", "push changes", "commit and sync", or "update remote". Handles the full workflow of committing changes, pulling with rebase, and pushing to remote.Votes: 0GitHub stars: 13
- CommitCommit current work with optional message. Handles git status/diff checking and message formatting.Votes: 0GitHub stars: 13
- Fast CliModern CLI tool mappings (eza, fd, rg, bat, dust, duf, fzf, zoxide, delta, jq). Use for file listing, searching, and disk usage.Votes: 0GitHub stars: 13
- Spec InterviewInterview-based spec development for large features. Iteratively questions user about implementation, UX, concerns, and tradeoffs before writing a complete specification.Votes: 0GitHub stars: 13
- Verification PlanningUse when about to verify, validate, or confirm anything in this repo — reproducibility checks, data validation, statistical analysis, integration or regression testing, or confirming a bug fix. Verification is a design problem, so plan how you'll verify before you start verifying.Votes: 0GitHub stars: 13
- Agent BrowserBrowser automation via agent-browser CLI: navigate, click, fill forms, screenshot, scrape, log in. Opens with a Chrome profile preflight.Votes: 0GitHub stars: 13
- Artifact WritingThe mechanics of publishing to the Artifact platform — md2artifact, the annotation layer (comments and suggested edits), rendering, republishing, and the viewer sandbox quirks. Use when publishing or updating any report, plan, spec, findings page, transcript view or dashboard as an Artifact, when running md2artifact, when building or debugging a select-to-comment layer, when a republish is refused, or when a control works locally but not in the viewer.Votes: 0GitHub stars: 13
- Artifacts SyncMaintain and reconcile a repo's ARTIFACTS.md — the index mapping artifact URLs to what they established. Covers the row schema, recording the publishing org, public-sharing limits under a Team/Enterprise org, and the reconcile pass. Use when publishing an artifact and recording it, when a link 404s, after switching Claude accounts or orgs, or on "sync/audit/refresh the artifact index".Votes: 0GitHub stars: 13
- BearRead and edit Bear notes via the Bear MCP (macOS only): search, create, edit, tag, archive, attach. Covers surgical edits and known failure modes.Votes: 0GitHub stars: 13
- CatalogA human-facing index of every skill and agent in this repo, grouped by category, with a one-line trigger for each. Use only when the user explicitly asks "what skills do I have", "what can you do", "is there a skill for X", or invokes `/catalog` directly — never fires on its own.Votes: 0GitHub stars: 13
- Check Bib ReferencesCatch LLM-fabricated citations in BibTeX files. Verifies arXiv/OpenReview entries against live metadata (titles, first authors), then guides manual verification of authorless prose claims. Use before submitting papers, after any LLM-assisted citation generation, or when a reference smells off.Votes: 0GitHub stars: 13
- Check Prose ClaimsFact-check prose claims — stats, comparatives, quotes — in slides, reports and PDFs. Use on 'fact-check this deck', 'audit this PDF'.Votes: 0GitHub stars: 13
- Chrome DevtoolsChrome DevTools MCP: evaluate JS, screenshots, network inspection, Lighthouse audits, performance traces, memory snapshots. Needs the extension.Votes: 0GitHub stars: 13
- Claude In ChromeClaude-in-Chrome MCP tools for your live browser tabs: read page content, navigate, fill forms, execute JS. For screenshots use chrome-devtools.Votes: 0GitHub stars: 13
- CouncilAsk another model family — one model, one agentic reviewer, one agentic reviewer per family, or the full eight-seat LLM council with blind peer ranking. Use for "second opinion", "sanity check this", "ask another model", "am I wrong", "llm council", "panel review", "review this with multiple models", "review this diff with several models", "adversarial review", before publishing a result or spec someone will act on.Votes: 0GitHub stars: 13
- Diagnosing BugsDiagnosis loop for hard bugs and performance regressions. Use when the user says "diagnose"/"debug this", or reports something broken/throwing/failing/slow. Prefer this over systematic-debugging for diagnosis — it forces a red-capable feedback loop and 3-5 ranked hypotheses before any fix is attempted.Votes: 0GitHub stars: 13
- Docs SearchFind where something is documented across a repo's Markdown — docs, specs, READMEs, CLAUDE.md and rules — without reading whole files into context. Use when asked "where is X documented", "do we have docs on Y", "what does the spec say about Z", when hunting for the convention or decision that governs a piece of code, or when a referenced doc's path is unknown and a plain Read has already missed.Votes: 0GitHub stars: 13
- DoneMark the session as done — generate a descriptive session title if it has none, prepend a checkmark, and hand back the /rename line that applies it. Use when wrapping up or closing out a session, and when the user says "done", "finished", "that's it", "ship it", "mark this done", or invokes /done.Votes: 0GitHub stars: 13
- House PlotsHouse chart style — pastel matplotlib defaults, palette, overlap checking. Use for any figure, chart or plot.Votes: 0GitHub stars: 13
- Inspect Ai EvalsSilent failure modes running inspect_ai evals against vLLM/Modal/RunPod endpoints: max_samples, max_connections, stop_reason, eval log.Votes: 0GitHub stars: 13
- Interview MeInterview the user relentlessly about a plan, decision, or idea to stress-test it. Use on "interview me", "grill me", "grill this plan", "poke holes in this", or "play devil's advocate."Votes: 0GitHub stars: 13
- JobsSubmit experiments or agent jobs with resource limits, check queue status, pause/resume workloads, troubleshoot slow machineVotes: 0GitHub stars: 13
- Llm JudgeBuild an LLM judge that scores text by meaning — intent, deception, tone — covering prompt design, blinding, fan-out, JSONL persistence.Votes: 0GitHub stars: 13
- Merge WorktreeMerge current worktree branch into the original branch, resolve conflicts with AI, then mark worktree for cleanup. Use when the user says "merge this worktree", "merge my branch back", or "finish this worktree."Votes: 0GitHub stars: 13
- ModalRun Python code in the cloud with serverless containers, GPUs, and autoscaling. Use when deploying ML models, running batch processing jobs, scheduling compute-intensive tasks, or serving APIs that require GPU acceleration or dynamic scaling.Votes: 0GitHub stars: 13
- Mv RepoMove or relocate a repo to a new directory — 'move this repo to <path>', 'relocate this project', 'move ~/code/X into ~/code/Y/'. Handles venv, Claude Code project state, symlinks, cache invalidation, path references, and tmux sessions.Votes: 0GitHub stars: 13
- Openrouter FusionQuery OpenRouter — a parallel multi-model panel with judge synthesis for deep-research and compare-and-contrast questions, a single non-Anthropic model for a one-off, or the raw HTTP reference for auth, the /models catalogue and the /chat/completions body. Use when reaching a non-Anthropic family, when a question wants several models answering in parallel rather than one, on "fusion", "ask GLM/Kimi/Qwen/DeepSeek", "what does another model think", or when writing code against the OpenRouter API.Votes: 0GitHub stars: 13
- Reduce AmbiguityReduce ambiguity in a draft before it is sent or acted on — specs, briefs, results pages, prompts. Use on "is this clear", "will this be misread", "red-team this doc".Votes: 0GitHub stars: 13
- Research LoopSet up and launch an autonomous overnight research loop — interview the user into concrete inputs, outputs and metrics, write a `program.md` spec, then dispatch the `autonomous-researcher` agent to run experiments unattended and review what came back in the morning. Use on "run experiments overnight", "automate this research", "set up an autonomous loop", "research loop", "run while I sleep", or whenever an agent is meant to iterate on a measurable metric with nobody in the room.Votes: 0GitHub stars: 13
- Results ArtifactStatistical machinery for a research results or findings page — which interval or test a number gets, the nulls a rate must be read against, chance correction and ceilings, and when to slice. Use when building or reviewing a results page, findings report or analysis writeup, when picking an interval or a significance test, when asking what a number would be with no signal, when correcting for chance or agreement, or when a pooled number spans strata that differ.Votes: 0GitHub stars: 13
- Server Storage TieringRelieve a full root disk on a server or cloud box with an attached volume: expose it at /workspace, tier data by access pattern, move never rm.Votes: 0GitHub stars: 13
- Setup ChannelSet up project-local messaging channels (Telegram, iMessage, Things Cloud) so Claude can send/receive messages and read tasks for this project. Use when the user asks to "set up Telegram", "connect iMessage", or "enable Things Cloud sync" for this repo.Votes: 0GitHub stars: 13
- ShipUse when the user says "ship this" or explicitly asks to review, finish, merge, and push implementation that is ready to deliver.Votes: 0GitHub stars: 13
- Skill Invocation ModesThe invocation-mode tradeoff when authoring a skill — model-invoked vs user-invoked, and the two loads you pay for the choice.Votes: 0GitHub stars: 13
- Spawn SessionSpawn a detached tmux session running Claude Code in a chosen directory, seeded with an initial prompt, optionally reachable by Remote Control.Votes: 0GitHub stars: 13