Visual ChangeNet for binary image classification and segmentation in AOI defect detection. Use when training, evaluating, exporting, or running inference for PCB defect detection or visual inspection, comparing image pairs for PASS/NO_PASS classification, or producing change-segmentation masks. Trigger phrases include "train Visual ChangeNet", "ChangeNet classify", "ChangeNet segment", "AOI defect detection", "PCB inspection model".

Multi-step video annotation pipeline that turns raw videos into Chain-of-Thought training data — multi-level captions, structured descriptions, and QA pairs (MCQ, binary, open-ended) with reasoning traces, via VLM/LLM distillation. Use when the user wants to "create video training data", "generate video QA datasets", "build CoT reasoning traces from videos", "auto-label videos", or run the video_reasoning_annotation pipeline. Triggers include "video annotation", "video CoT", "video QA", "chain-of-thought", "video captioning pipeline", "video distillation".

Fine-tune any HuggingFace CV / VLM / LLM model on local NVIDIA GPUs inside an NGC PyTorch container when no dedicated TAO model skill matches. Use when the user wants to fine-tune a HuggingFace model (full or LoRA), train a vision / VLM / LLM model end-to-end, generate a reproducible HF training pipeline, smoke-test a HuggingFace model locally before scale-up, push a fine-tuned model to the HF Hub with a model card, or emit a self-contained rerun skill for an existing HuggingFace finetune. Supports image classification, object detection, semantic / instance / panoptic segmentation, depth estimation, image-text-to-text VLM (SFT / LoRA), and LLM SFT / DPO / GRPO. Six-step workflow: inspect and qualify, hardware and NGC image, research, generate and smoke, train + eval + infer, push and emit rerun skill. Do not use for any Hugging Face model ID claimed by a dedicated `skills/models/*` skill; the model skill and its declared execution environment take precedence.

Use when searching, extracting, ingesting, or querying a document collection with the NeMo Retriever 26.8.1 CLI, including local LanceDB indexes and deployed Retriever services. Use for PDFs, images, Office files, HTML, text, audio, and video; not for editing documents or web search.

Fetch hot finance news, unified trends, and prediction financial market data. Use when the user needs real-time financial news, trend reports from multiple finance sources (Weibo, Zhihu, WallstreetCN, etc.), or Polymarket finance market prediction data.

This skill should be used when generating comprehensive test cases from PRD documents or user requirements. Triggers when users request test case generation, QA planning, test scenario creation, or need structured test documentation. Produces detailed test cases covering functional, edge case, error handling, and state transition scenarios.

Assigns cell-type labels to clusters from a protein marker panel (multiplexed imaging such as MIBI, CODEX, IMC and CyCIF; mass and flow cytometry; CITE-seq protein), where the labels are judged against an expert reference. Within-dataset normalisation of cluster summaries, lineage first by positive defining markers, subtype only as far as the panel can determine it, marker-poor clusters assigned by exclusion, class-coverage sanity checks, and two independent annotators with adjudication. Use when the output is one label per cluster and the grader is agreement with an expert; for gating single cells in FCS data use flow-cytometry-analysis, and for transcript-based annotation use scanpy or scvi-tools.

Discover, find, compare, audit, repair, adapt, craft, run, debrief, save, and prepare repeatable AI-agent loops for publication. Use when a user asks to analyze code or coding threads for recurring work, find a published loop, interview them to turn a goal into a bounded loop, review a loop for weak checks or unsafe authority, execute a loop with an evidence receipt, learn from completed runs, save or reuse a project loop, or validate and submit a loop to Loop Library.

Health-check the wiki for contradictions, stale claims, orphan pages, missing cross-references, and knowledge gaps. Use periodically or when the user says 'lint the wiki' or 'check wiki health'.

Write 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.

Sweep open (or listed) PRs with up to 100 parallel agents: security-scan outside contributors, rebase onto main when behind (push --force-with-lease), approve pending first-time-contributor CI when relevant, optionally rebase in-house PRs, and report who should review. Supports full, changed-only, behind-only, and conflict-only scopes for cheap daily runs. Use when the user runs /pr-sweep (or /pr-inbound-sweep), or asks to "sweep PRs", "sweep inbound PRs", "security-check outside PRs", "rebase outsider PRs", "rebase our PRs", "approve waiting CI on PRs", "daily PR sweep", or "prep external PRs for review".

Triage all open GitHub issues, PRs, and discussions in the current repository by fanning out up to 100 parallel subagents (one per item), then produce a single prioritized report ranking which PRs to review first, which issues to address first, and which discussions need maintainer attention. Use when the user asks to "triage open issues/PRs", "triage discussions", "prioritize the backlog", "what should I review first", "sweep the repo", or any request to bulk-evaluate open GitHub work and recommend an order.

Finds, reads, and reconstructs coding-agent sessions across Codex, Claude, OpenCode, OMO/Senpi, and other local agent logs. Use when asked to find or search past sessions, transcripts, or subagent runs, or to recover what an earlier session did.

Configures a language server so editor/agent tooling (diagnostics, go-to-definition, references, rename) works. Use when a project needs an LSP installed or wired, or a 'no LSP server configured' error appears.

Use when finding, comparing, reading, or monitoring arXiv papers, including requests for abstracts, citation graphs, original LaTeX, section-level technical details, or literature reviews.

Calculate an ETF's premium or discount to NAV from Yahoo Finance data (yfinance), compare or screen ETFs by premium, explain why a gap exists, and decompose a sudden ETF move into NAV-driven vs structural components (dealer gamma exposure, blocked AP arbitrage, sentiment). Use this skill whenever the user asks whether an ETF trades above or below NAV, compares ETF premiums or discounts, screens for the biggest ones, asks about ETF arbitrage or premium convergence, or wants to know why an ETF jumped or diverged from its holdings — including gamma squeezes, dealer gamma exposure (GEX), and blocked creation/redemption. Especially relevant for leveraged, inverse, international, bond, commodity, and crypto ETFs (IBIT, BITO, HYG, KWEB).

Analyze a company's most recent (or a specified past) earnings report from Yahoo Finance data (yfinance): actual vs estimated EPS, surprise size, revenue and margin trends, and the stock's price reaction. Use this skill whenever the user asks how earnings went — beat or miss, earnings surprise, quarterly results, the post-earnings move, or an earnings call recap — including casual references to a past report such as "AMZN reported last night" or "how did they do". For an upcoming report, use earnings-preview.

Estimate a public company's intrinsic value with DCF, relative (peer multiple), and sum-of-the-parts (SOTP) methods, then blend them into an implied share price with upside/downside vs the market price, a WACC and terminal-growth sensitivity grid, and bull/base/bear scenarios. Use this skill whenever the user asks what a company or ticker is worth: fair value, intrinsic value, implied share price, a price target from fundamentals, whether it is overvalued or undervalued, building a DCF (WACC, terminal value, discounted cash flow), EV/EBITDA or P/E based targets, peer comparison valuation, or SOTP and conglomerate discounts. Run the model rather than answering valuation questions from memory.

Validate a commit-specific Next.js preview package and manually trigger the entire Next.js deployment test suite through the test_e2e_deploy_release.yml GitHub Actions workflow. Use only when asked to run the full deploy test suite or this workflow specifically from an internal vercel/next.js PR branch. Do not use for focused deployment-test sanity checks; run the relevant tests locally with pnpm test-deploy instead. Covers resolving the latest branch SHA, waiting for vercel-packages, preserving default workflow inputs, dispatching the workflow, and verifying the run.

Scope a feature end to end and write an implementation spec under specs/ from an accepted proposal — current-system mapping, goals/non-goals, a Decision Log for one-way-door choices, a live Open Questions backlog, and a real migration + test plan. Read when asked to write a spec, scope this feature, turn this proposal into a spec, plan the implementation, or break this into tasks. Do NOT fire on frame a proposal or write the PRD (sibling frame-a-proposal — a PRD frames a change before it is accepted; this skill starts once one is), record a decision or write the ADR (record-a-decision), write a postmortem (write-a-postmortem), or review this design (review-a-design) — those are separate skills. Complements the platform open-knowledge skill; does not replace it.