Generates a personal markdown changelog of merged or closed pull requests authored by the current user and Linear tickets the user closed or worked on, over a configurable window (default 7 days), grouped by feature area (e.g. Dashboards, Agent0). Inputs sourced from `gh search prs --author=@me` and the Linear MCP. Use for weekly recaps, status updates, performance reviews, or end-of-sprint summaries. Triggers on "changelog", "what have I done", "weekly summary", "my recent work", "recap my week", "/changelog".

Iterates on a failing AI/LLM eval (an L2 suite, a golden-set / judge eval, or any eval gating a PR) until it is green AND confirmed, not just luckily passing once. Classifies the failure as a code bug, an eval-definition bug (stale golden item / criteria drift), judge-drift (silent grader version change), or flaky. Applies the minimal fix, then requires N consecutive confirming re-runs — 2 deterministic, 5 for anything a model call grades, since a flat "2" is statistically weak for a stochastic judge. Hard-capped at 5 fix iterations. Refuses to game the eval (Goodhart's Law: skip/delete/overwrite-in-place a case, loosen a threshold) without a second independent check plus `confidence(analysis) >= 90%` and a logged rationale. Composes `ai-engineering`, `confidence`, `critical`, `verify-behavior`. Use when an eval is failing on a PR and needs a real, non-gamed green. Triggers on "this eval is failing", "iterate on this eval", "fix this eval", "get this eval green", "optimize this eval", "/eval-iterate".

Writes rigorous mathematical proofs for ML/AI theory. Use when asked to prove a theorem, lemma, proposition, or corollary, fill in missing proof steps, formalize a proof sketch, 补全证明, 写证明, 证明某个命题, or determine whether a claimed proof can actually be completed under the stated assumptions.

Judge and shape UI/UX for products used by parents and young children - audience fit, accessibility, safety and trust, comfort, consistency, and human feel. Use when reviewing or designing family, parenting, or kids app screens on phone or tablet.

Maintain a brand kit — the canonical brand context an ad or content pipeline reads (positioning, audience, voice, standing instructions, brand-type, value-props, colors), plus manage the product list and attach product photos. Use when someone says "update my brand kit", "set my brand voice/audience", "add a product to my brand", or hands you a folder of product shots to attach. Platform-agnostic: it teaches the field model, partial-update/clear semantics, override behavior, caps and validation, and the hero-image and product-matching rules — independent of any specific backend.

Write tests, mock functions, assert values, and configure test suites for Roblox/Luau code using Jest Roblox — a Luau port whose API diverges from JS Jest in subtle, non-obvious ways that LLM training data does not cover. Use whenever encountering Jest in Roblox/Luau code: editing .spec test files, calling jest.fn/jest.mock/spyOn, using describe/it/expect, or debugging Jest tests.

Fit, summarize, plot, and interpret a chosen CausalPy experiment. Use after the causal method has been selected, including when configuring PyMC/sklearn models and scale-aware custom priors.

Implements authentication, authorization, encryption, secrets management, and security hardening patterns. Use when designing auth flows, managing secrets, configuring CORS, implementing rate limiting, or when asked about JWT, OAuth, password hashing, API keys, RBAC, or security best practices.

Trace DSPy programs and ReAct agents with Maple: OpenInference DSPy instrumentor with GenAI dual-write, a DSPy callback for tokens, cost, tool names and agent spans, using_session for one session per conversation, and thread context for dspy.Parallel. Triggers on 'trace my dspy agent', 'add Maple to dspy', 'agent sessions for dspy', 'OpenTelemetry for dspy'.

Retrieve chemical compound data from PubChem and ChEMBL with disambiguation, cross-referencing, and stereochemistry handling. Use for resolving compound names to SMILES/InChI/CID/ChEMBL IDs (including OPSIN deterministic IUPAC-name-to-structure parsing), fetching molecular properties, distinguishing isomers/stereo forms, and cross-validating identity across databases. Always use English compound names; flags ambiguous queries (e.g., Vitamin D has multiple forms).

Deep literature review — PubMed, EuropePMC, bioRxiv preprints, citation networks, evidence synthesis. Disambiguates queries, runs collision-aware searches, grades evidence T1-T4, and produces structured reports. Use for systematic literature review, meta-analysis evidence collection, and detailed answer-with-citations workflows.

Install and configure ToolUniverse for any use case — MCP server (chat-based), CLI (command line with 14 subcommands), or Python SDK (Coding API with 3 calling patterns). Covers uv/uvx setup, MCP configuration for 12+ AI clients (Cursor, Claude Desktop, Windsurf, VS Code, Codex, Gemini CLI, Trae, Cline, etc.), full CLI reference (tu list/grep/info/find/run/test/status/build/remote/doctor/serve/connect/connections/disconnect), Coding API quickstart, agentic tools, code executor, API key walkthrough, skill installation, and upgrading. Use when user asks how to set up ToolUniverse, which access mode to use (MCP vs CLI vs SDK), configuring MCP servers, using the CLI, troubleshooting installation, upgrading, or mentions installing ToolUniverse or setting up scientific tools. Also triggers for "how do I use ToolUniverse", "what's the best way to access tools", "command line", "tu command", "coding API", "tu build".

Quickly triages a contributor pull request or reported issue against repository policy and lightweight evidence without full review or root-cause analysis. Publishes the disposition to GitHub, and on a contributor pull request that links no issue it may open a tracking issue and close the pull request. Use when the user asks to "triage this contributor PR", "sanity-check this PR", "triage this issue", "reproduce this reported issue", "check whether this bug still exists", or invokes /prp-maintainer-triage.

Autonomously owns one workstream from an issue, PRD, document, existing plan, or free-form request through planning, implementation, pull request, independent review, corrections, and green CI. Always use when the user asks to implement or ship work end to end, take an issue or idea to a reviewed PR, run plan to PR, invokes /prp-issue, or when prp-orchestrate needs an end-to-end delivery engine.

Extract clean markdown content from web pages using Defuddle CLI, removing clutter and navigation to save tokens. Use instead of WebFetch when the user provides a URL to read or analyze, for online documentation, articles, blog posts, or any standard web page.

Attach an image or video that already exists to a GitHub pull request or issue with `gh --attach`, in the description or as a comment. Use for "attach this image to the PR", "put this file in the issue", "put the image under Evidence in the open PR". To take or annotate the image first, use `visuals`.

Use when building a weighted gene co-expression network from a bulk expression matrix and a sample group file, filtering variable genes by MAD, identifying co-expression modules with WGCNA, correlating modules with traits, and exporting module-level plots and gene tables. NOT for single-cell RNA-seq, differential expression testing, methylation analysis, or datasets that are too small for WGCNA after quality control.

Generates complete Mendelian Randomization + single-cell transcriptomics (scRNA-seq) research designs from a user-provided direction. Always use this skill whenever a user wants to design, plan, or build a study combining MR and single-cell data — even if phrased as "help me write a paper on X", "design a bioinformatics study for Y", or "I want to study Z using MR and scRNA". Covers five study patterns (mechanism gene-set, key-cell, candidate-gene reverse validation, exposure-disease-cell triangulation, translational biomarker) and always outputs four workload configs (Lite / Standard / Advanced / Publication+) with recommended primary plan, step-by-step workflow, figure plan, validation strategy, minimal executable version, publication upgrade path, and a strictly verified reference literature retrieval layer with real references only.

A medical-research-native literature reading skill for users with clinical, bioinformatics, translational, and basic experimental backgrounds. Use this skill whenever a user wants to read, analyze, critique, or interpret a medical or scientific paper — whether they provide a PDF, abstract, DOI, PMID, or just a title. Triggers include requests like \"analyze this paper\", \"critique this study\", \"is this a strong paper?\", \"give me similar studies\", \"prepare me for journal club\", \"help me understand this bioinformatics paper\", \"what are the weaknesses here?\", or \"turn this into a mind map\". Also activate for any downstream deliverables such as journal club kits, comparison tables, PI decision briefs, replication starters, or follow-up experiment designs. Do NOT treat as a generic summarizer — this skill performs structured evidence-type classification, track-specific critical appraisal, interpretation-boundary judgment, and research-grade follow-up generation.

Run a submission-readiness preflight on a manuscript before arXiv upload. Use when the user is preparing an arXiv submission, asks to check a paper before uploading, mentions hallucinated or fake references, leftover LLM meta-comments / prompts in text, placeholder data (TODO, TBD, XX%), AI-use disclosure, scholarly integrity, research integrity, or arXiv moderation risk — even if they don't say "preflight". Also trigger on phrases like "check my paper before arXiv", "verify my references", "scan for AI artifacts", "scan for LLM residue", "is my submission ready", or "review .tex/.bib before submit".