maragudk
- 35 skills
- 0 followers
- 6 hours ago last updated
- ▌ LLM As A Judge · maragudk bundleBuild, validate, and deploy LLM-as-Judge evaluators for automated quality assessment of LLM pipeline outputs. Use this skill whenever the user wants to: create an automated evaluator for subjective or nuanced failure modes, write a judge prompt for Pass/Fail assessment, split labeled data for judge development, measure judge alignment (TPR/TNR), estimate true success rates with bias correction, or set up CI evaluation pipelines. Also trigger when the user mentions "judge prompt", "automated eval", "LLM evaluator", "grading prompt", "alignment metrics", "true positive rate", or wants to move from manual trace review to automated evaluation. This skill covers the full lifecycle: prompt design → data splitting → iterative refinement → success rate estimation.
- ▌ Trace Annotation Tool · maragudk bundleGenerate a custom trace annotation web app for open coding during LLM error analysis. Use when the user wants to review LLM traces, annotate failures with freeform comments, and do first-pass qualitative labeling (open coding). Also use when the user mentions "annotate traces", "trace review tool", "open coding tool", "label traces", "build an annotation interface", "review LLM outputs", or wants to manually inspect pipeline traces before building a failure taxonomy. This skill produces a tailored Python web application using FastHTML, TailwindCSS, and HTMX.
- ▌ Observability · maragudk bundleGuide for instrumenting and operating observable software systems. Use this skill whenever the user is adding or reviewing telemetry, instrumenting code with OpenTelemetry, working with traces/spans/metrics/logs/structured events, debugging production behavior, designing SLOs or alerts, setting up sampling or telemetry pipelines, choosing observability storage, or making code observable for humans or AI agents -- even if they don't say the word "observability". Triggers include "add tracing", "instrument this", "OTel/OpenTelemetry", "why is this slow in prod", "set up an SLO", "alert fatigue", "high cardinality", "structured logging", "wide events", and observability for LLM/agent applications.
- ▌ One At A Time · maragudkPresent output one unit at a time instead of dumping a list or wall of text. Use when the user says "one at a time", "/one-at-a-time", "one thing at a time", "go one at a time", asks you to slow down, to stop the wall of text, or to stop dumping everything at once -- either to redo output you just gave, or to pace the rest of a task.
- ▌ Second Opinion · maragudkGet a second opinion from another AI model (currently OpenAI models, via the codex CLI) on a design decision, a stubborn bug, a code review, or any question where an independent take is valuable. Use only when the user explicitly asks for one -- "get a second opinion", "ask OpenAI", "ask codex", "what does GPT think", "/second-opinion", or similar. Do not use proactively.
- ▌ Simple English · maragudk bundleWrite or rewrite technical text with the rules of ASD-STE100 Simplified Technical English so it is clear, unambiguous, and free of AI slop. Use for documentation, READMEs, runbooks, procedures, error messages, release notes, incident reports, and API guides. Also use when the user says "STE", "Simplified Technical English", "ASD-STE100", "de-slop", "make this readable", "write for non-native readers", or asks for docs that translate well. Enforces the standard's 53 rules: 20/25-word sentence limits, one word one meaning, simple tenses, active voice, condition before command.
- ▌ Accomplishments · maragudkExtract accomplishments from new implementation diary entries into docs/accomplishments.md, written as one-line reflections that blend self-appraisal with Markus's appreciation. Use only when the user invokes /accomplishments or explicitly asks to extract, update, or refresh accomplishments from the diary. Do not run proactively.
- ▌ Observable Plot · maragudk bundleGuide for using Observable Plot, a JavaScript library for exploratory data visualization built on D3. Use this skill when creating charts, exploring data, or building visualizations with marks, scales, and transforms.
- ▌ Security Review · maragudk bundlePerform a thorough security review of the project, starting from a randomly selected file. Use this skill when the user asks for a security review, security audit, vulnerability scan, or wants to check the codebase for security issues. Also trigger when the user mentions "check for vulnerabilities", "find security bugs", "OWASP", or any request related to assessing the security posture of the project.
- ▌ Blog Post Interview · maragudkInterview the user about a new blog post before writing it. Use this skill whenever the user wants to write a blog post, says "let's write a post about X", "I want to blog about Y", "help me draft a post on Z", or otherwise asks for help producing a blog post. The skill conducts the interview; you write the post afterward based on what you learn.
- ▌ Writing Clearly And Concisely · maragudk bundleUse when you write or revise a long-form document an audience will read—documentation, a README, a how-to or user guide, a spec, a design or decision doc, a blog post, a proposal, an announcement, or a standalone report or summary written for a person. Such writing should be clear, concrete, and tight, so draft it from the start by Strunk's Elements of Style (active voice; definite, specific, concrete words; omit needless words) instead of first-draft sprawl. Do NOT fire for short or code-adjacent text—commit messages, PR descriptions, release notes or changelogs, error messages, UI strings, code comments, config, version bumps, quick internal scratch notes, or the implementation diary. These exclusions win even when the text reaches an audience: a release note, changelog, or PR description is a working artifact, not a document drafted for readers.
- ▌ Go · maragudkALWAYS invoke this skill as your FIRST action — before any Read, Write, Edit, or Bash — whenever a task involves Go in any way — creating or editing any `.go` file, `go.mod`, or `go test`; adding or changing handlers, structs, methods, interfaces, error handling, goroutines, channels, sqlc/database code, or gomponents views; or refactoring, reviewing, debugging, or testing Go. This is a hard requirement, not a suggestion — the user develops all Go through this skill, which encodes project-specific conventions (structure, error handling, testing, naming) that default Go habits violate. Do not start editing Go and consult it later — load it first. Skip only when no Go code is read or written (pure Python, shell, SQL, Dockerfiles, CI config, or conceptual Go questions with no code).
- ▌ Git · maragudkGuide for using git with specific preferences -- backticks around code identifiers in commit messages, asking about GitHub issues to reference before committing. Use this whenever you commit, write a commit message, or work with pull requests -- not just when explicitly asked to "commit". These conventions aren't in your default knowledge and you'll get them wrong without consulting this skill.
- ▌ Xbrl · maragudk bundleGuide for working with XBRL, the standard for digital business and financial reporting -- reading and writing instance documents, Inline XBRL (iXBRL), xBRL-JSON and xBRL-CSV, taxonomies and linkbases, dimensions, contexts, units, and facts. Use this skill whenever the user touches XBRL in any form -- parsing or generating an instance, extracting figures from a filing or annual report, handling `xbrli:context`, `contextRef`, `unitRef`, `decimals`/`precision`, `ix:nonFraction`, `ix:nonNumeric`, `ix:hidden`, `xbrldi:explicitMember`, `xbrldi:typedMember`, hypercubes, DTS discovery, taxonomy or report packages, or the Open Information Model. Triggers even when the word "XBRL" never appears -- ESEF, iXBRL, SEC EDGAR company facts, `us-gaap:`/`ifrs-full:`/`dei:` prefixed tags, financial statement tagging, or a file arriving as `.xbrl`/`.xbri`/`.xbr` are all enough.
- ▌ Modal · maragudkGuide for running Python code on Modal, the serverless compute platform for AI workloads, batch jobs, scheduled tasks, web endpoints, and sandboxed code execution. Use this skill whenever the user is writing or modifying Modal code (anything importing `modal`, decorating with `@app.function`, `@app.cls`, `@modal.fastapi_endpoint`, etc.), running `modal run`/`modal deploy`/`modal serve`, configuring GPUs/images/volumes/secrets for Modal, or asking how to host inference, fine-tuning, or agent sandboxes on Modal.
- ▌
- ▌ Atproto · maragudk bundleGuide for building on the AT Protocol (the "atmosphere") -- authoring Lexicons, building app views, consuming the firehose, working with identity (DIDs, handles), repositories, records, XRPC endpoints, and OAuth. Use this skill whenever the user is building anything on atproto/Bluesky/the atmosphere -- writing Lexicon JSON, calling com.atproto.* or app.bsky.* endpoints, parsing AT URIs (`at://...`), DIDs (`did:plc:...`, `did:web:...`), handles, TIDs, the indigo Go SDK (`github.com/bluesky-social/indigo`), the firehose / `subscribeRepos`, MSTs, CAR files, DAG-CBOR/DRISL, app views, feed generators, labelers, or PDS interactions. Triggers even if the user doesn't say "atproto" -- words like "lexicon", "PDS", "app view", "firehose", "did:plc", or `at://` URIs are enough.
- ▌ Bluesky · maragudkGuide for posting content to the Bluesky social network using the bsky terminal app. This skill should be used proactively when working in public repositories and there is interesting, shareable content (new features, insights, achievements, or announcements worth sharing with the community). Use it when asked to post to Bluesky, or when content seems worth sharing publicly.
- ▌ Unsloth · maragudk bundleGuide for fine-tuning LLMs, embedding models, vision-language models, and TTS models efficiently with Unsloth. Covers LoRA/QLoRA SFT, reinforcement learning (GRPO, DPO, ORPO, KTO), embedding fine-tuning with sentence-transformers, continued pretraining, and saving/exporting to GGUF, Ollama, or vLLM. Use this skill whenever the user mentions Unsloth, FastLanguageModel, FastSentenceTransformer, FastVisionModel, FastModel, or wants memory-efficient fine-tuning of open LLMs or embedding models on a single GPU, even if they don't explicitly say "Unsloth".
- ▌ Dad Joke · maragudkTell the user a dad joke. Use this skill when the user asks for a dad joke, wants to hear something funny, needs cheering up, or says /dad-joke.
- ▌ Datastar · maragudkGuide for building interactive web UIs with Datastar and gomponents-datastar. Use this skill when adding frontend interactivity to Go web applications with Datastar attributes.
- ▌ Decisions · maragudkGuide for recording high-level architectural, design, and product decisions in docs/decisions.md. This is NOT a log of every choice made during a regular build -- those belong in the implementation diary. Use this skill only when a clearly significant decision is made (database choices, frameworks, core design patterns, product direction) or when explicitly asked to document a decision. Also suggest proactively at natural session-end moments -- after a PR merges, a feature ships, or a work chunk wraps up -- if a significant decision was made during the session and not yet recorded. Be conservative - only suggest for major decisions, not minor implementation details.
- ▌ Gardeners · maragudkAutonomous project gardening by a coordinated team of agents. Spawns a team of gardeners that each run the `garden` skill in parallel, coordinating via a shared task list to avoid duplicate work. Use when the user wants to tend multiple small issues in one pass. Invoke with /gardeners.
- ▌
- ▌ Gomponents · maragudkBuilding, rendering, or editing any HTML or UI in a Go application means using gomponents — ALWAYS invoke this skill as your FIRST action, before any Read, Write, Edit, or Bash. This fires for any view, page, layout, component, form, table, navbar, footer, or list that renders to HTML; any function returning a `Node`; anything in the `html` package; and converting HTML or templates into Go. It is a hard requirement — the user writes all HTML through gomponents, a pure-Go component library whose conventions (dot imports, `Node` composition, `Map`/`If`/`Iff`, `Group`, HTML5 documents) ordinary Go and HTML habits get wrong. Load it even when the Go skill also applies. Skip only when no HTML or view code is touched (pure database, handler, or business logic; plain `.css`/`.js`/`.html` files; conceptual questions with no code).
- ▌ Nanobanana · maragudkGuide for generating and editing images using generative AI with the nanobanana CLI
- ▌ Code Review · maragudkGuide for making code reviews. Use this when asked to make code reviews, or ask to use it before committing changes.
- ▌ Turbopuffer · maragudkGuide for building search on turbopuffer, the object-storage-native vector and full-text search engine. Use this skill whenever the user is writing or modifying code that talks to turbopuffer -- anything importing `github.com/turbopuffer/turbopuffer-go`, calling `*.turbopuffer.com` endpoints, or mentioning tpuf or turbopuffer namespaces -- and when designing features it could serve, such as semantic/vector search, BM25 full-text search, hybrid search with RRF, or RAG retrieval with filters. Also use it when deciding whether turbopuffer fits a use case, designing namespaces or schemas for it, or debugging its consistency, caching, backpressure, or HTTP 429 behavior.
- ▌ Distill Book · maragudkDistill a long book into a concise, structured set of learnings by processing it chapter by chapter with parallel subagents, then synthesizing the result -- optionally into a Claude Code skill. Works for any format the book comes in -- PDF, EPUB, Markdown, HTML, plain text. Use this skill whenever the user wants to summarize, distill, extract the core ideas from, or "turn into a skill" a book, ebook, manual, or other long-form document that is too large to read in one pass. Triggers include "distill this book", "summarize this book chapter by chapter", "extract the key principles from this PDF/ebook", "make a skill out of this book", or handing over a large multi-chapter document and asking what to learn from it.
- ▌ Macos Design · maragudk bundleDesign and build native-feeling macOS application UIs. Use this skill whenever the user asks to create a desktop app, macOS app, Mac-style interface, Apple-style UI, system utility, or anything that should look and feel like a native Mac application. Also trigger when users mention "native feel", "desktop app design", "Apple design patterns", "sidebar layout", "traffic lights", or want to build tools/utilities that feel like they belong on macOS. This skill covers layout, composition, interaction patterns, animations, light/dark mode, and all the subtle details that make an app feel like Apple built it.
- ▌ Improve Skill · maragudkReview the current conversation for fabrik skills that could be improved (corrections, friction the user had to manually flag, missed triggers, anything else worth flagging) and ship the improvements back to the fabrik repo as a PR (concrete fixes) or issue (fuzzy observations / redesigns). Use when the user invokes /improve-skill or asks to make a skill better, smarter, or less friction-prone. May also be suggested at end of session if there's concrete signal that a skill underperformed; otherwise stay silent.
- ▌ Diary · maragudkWrite and maintain an implementation diary capturing what changed, why, what worked, what failed (with exact errors and commands), what was tricky, and how to review and validate. Activates proactively during non-trivial implementation work (new features, bug fixes, refactors, research spikes). Does not activate for trivial tasks like one-line fixes, config tweaks, or quick questions.
- ▌ Garden · maragudk bundleAutonomous project gardening. Scans for maintenance issues (starting with documentation), picks one, fixes it in a worktree, self-reviews with competing agents, and opens a PR. Use when the user wants to tidy up the project, fix stale docs, or generally tend the codebase. Invoke with /garden.
- ▌ Autoresearch · maragudkAutonomous experiment loop that iteratively improves a measurable metric. Given a goal, a verify command, and an optional guard, the agent branches, makes one change, measures the result, and keeps or discards the experiment -- repeating indefinitely. Use this skill when the user wants to optimize something measurable through automated experimentation, autonomous improvement loops, or when they mention "autoresearch". Works for any domain with a quantifiable metric (code performance, ML training, build size, test scores, content quality metrics, etc.).
- ▌ Address Code Review · maragudkAddress code review feedback by walking through comments one at a time with the user. Use when the user has received code review comments — on a GitHub PR, in a document in the repo, or directly in conversation — and wants to work through them methodically. Also trigger when the user mentions "address review", "review comments", "PR feedback", or wants to respond to code review feedback.