Art & Design

Use when an image needs text extraction via OCR.

Use when bringing a recipe in the spark-recipes repo up on the 2-node DGX Spark cluster — starting/staging a recipe directory, tearing down the currently-serving model, rebuilding the patched image, launching worker-then-head, and confirming health. Triggers: "bring up <recipe>", "start <recipe> on the sparks", "deploy the glm/qwen/deepseek recipe", "latest recipe not running", "fresh head/worker after a model change", DFlash2/DFlash enablement.

Extract, clean, join, update, verify, and deliver Excel, CSV, cloud spreadsheets, or tabular data from images through a conversation-driven agent workflow. Use when the user asks an agent to 做表格、合并表格、图片转Excel、清洗数据、更新Excel、Google Sheets、飞书表格 or WPS表格.

Search and fetch royalty-free stock photos from the Pexels API. Use when the user asks to find stock photos, seed placeholder/demo/seed images, populate a gallery, or needs real royalty-free imagery for a project (e.g. "find me 5 photos of X", "get stock images for the hero section", "seed images from Pexels").

Implement or refine a UI component from a reference screenshot by harnessing image comparison with the pxp CLI. Use for requests like "build this component from this PNG", "make this component match", or "pixel perfect this UI" when application code changes are expected. Also supports screenshot comparison without code changes when explicitly requested. Not for implementing a UI without a visual reference.

Create, revise, audit, and export submission-grade scientific figures for Nature-family and other high-impact venues in Python (matplotlib/seaborn) or R (ggplot2/patchwork/ComplexHeatmap), including multi-panel plots, figures4papers-style work, and journal-ready SVG/PDF/TIFF outputs. Use for paper or scientific plots, manuscript data visualization, 论文配图、学术写作配图、科研绘图、科研作图、画图、作图、出图、论文图表、可视化. Define the conclusion, evidence logic, data integrity, template compatibility, export needs, and reviewer risks before plotting; honor or persist the Python/R backend choice. Also use the separate OpenRouter GPT Image 2 route for explicit AI-generated graphical abstracts, mechanism diagrams, concept schematics, 论文示意图、机制示意图、图形摘要; this route skips backend choice and treats outputs as drafts. Do not use for interactive dashboards, statistics-only analysis, data cleaning, literature review, code debugging, pure photo editing, or Illustrator/Figma-first infographics without manuscript-figure intent.

Generate or edit product, editorial, photographic and typography images with GPT Image 2.5. Sunburst favors quality; Flare favors speed. Automatically use the other variant on rate limits.

Higgsfield image prompt director covering Banana Pro, Soul Cinema, and GPT-2. Three asset types in strict order: (1) single-image character outfit on white seamless studio — the locked base reference, via Banana Pro (custom styling from prompt) or Soul Cinema (face/body from character ref + outfit from wardrobe ref), (2) 6-panel multi-angle character sheet off that base, (3) scene plates with or without characters in cinematic environments. Also GPT-2 for detail face/chest-up portraits. Reads reference images for hair, makeup, wardrobe, jewelry, identity markers. Outputs photorealistic prompts with locked hyperreal stack — skin pores, subsurface scattering, strand-by-strand hair, fabric weave, Kodak Vision3 film. Use for any character outfit reference, character sheet, model sheet, ref sheet, multi-angle sheet, scene plate, environment plate, face detail shot, or photorealistic still — even without saying 'Banana Pro,' 'Nano Banana,' or 'Soul Cinema.'

Use when implementing or reviewing iOS app-wide error handling where SwiftUI views/ViewModels notify common errors and an @main App, WindowGroup AppShell, NavigationStack coordinator, or UIKit AppCoordinator owns root navigation, tab/session flow switching, alerts, sheets, toast hosts, SessionExpired handling, Maintenance handling, and root reset behavior.

Extract every dated or triggerable obligation from an executed contract into an oblig-register YAML file (schema in this repo), including the notice clause as deemed-receipt parameters. Then validate, resolve deadlines, and export a calendar. Use when someone asks "what do we have to do under this contract and when", "put this contract's dates in the calendar", or "build an obligations register".

Reads, searches, and exports local Claude Code history without resuming work. Covers recent session inventory, exact session timelines, verbatim human input including queued mid-turn prompts, full-event keyword search, hybrid recall when wording changed, end-state triage, and deleted-file recovery across active Claude homes plus registered archives. Use whenever the user asks what they or Claude said, wants a Claude Code session ID or original context, remembers prior work vaguely, needs an old file from a transcript, or must prove what a Claude session contained before continuing it. Also owns the only Kimi CLI surface, via its Kimi inventory and search flags. For Codex history use read-codex-history; when the request names no platform at all or spans providers, start at local-conversation-history.

Flutter implementation and review checklist. Use only when writing, modifying, or reviewing Flutter widget code, layout constraints, adaptive sizing, navigation and routing, platform channels, platform views, background entry points, list and image performance, async gaps across `BuildContext`, disposal, theme and design tokens, accessibility, localization, offline and error states, or Flutter test and golden coverage. Do not use for Dart language generalities, native Android or iOS internals beyond the engine and channel-registration seam a Dart entry point depends on, or broad rewrites.

Build a litigation chronology from documents (correspondence, affidavits, pleadings, contracts) as chronology-schema YAML with a pincite on every entry; then merge sources, flag date conflicts and evidential gaps, and render. Use when asked for "a chronology", "timeline of events", "what happened when", or to reconcile two accounts of events.

Coding Agent Session Search - unified CLI/TUI to index and search local coding agent history from Claude Code, Codex, Gemini, Cursor, Aider, ChatGPT, Pi-Agent, Factory, and more. Purpose-built for AI agent consumption with robot mode.

Converts books and documents (PDF, EPUB, DOCX, HTML, Markdown, plain text, RTF, MOBI/AZW with Calibre) into structured agent skills, extracting frameworks, mental models, principles, techniques, and anti-patterns. Use when the user wants to study a document through GitHub Copilot CLI, Amp, Claude Code, or Hermes Agent, apply an author's frameworks while working, or build a reusable knowledge base from a file.

Reads, searches, and exports local OpenAI Codex history without continuing the old task. Lists recent Codex sessions, extracts exact prompt-ledger inputs by Session, locates a rollout by verified session_meta identity, reconstructs one chronological user/assistant timeline with fork and compaction lineage, and performs bounded keyword search across live and archived rollouts. Use whenever the user asks what they told Codex, wants recent original inputs, a Codex Session ID, full prior context, fork ancestry, or evidence of what a Codex run did. For Claude Code history use read-claude-code-history.

Non-interactive dependency + agent-skill refresh across a fleet of web repos — update to latest under a pinned supply-chain release-age cutoff, re-lock, prune, audit, dedupe-dry-run, refresh vendored agent skills, then run each repo's quality suite. Flags skills that failed because they were renamed or removed upstream and reconciles them. Reports a per-repo table and a cross-fleet common-dependency drift matrix. Leaves changes uncommitted.

Autonomous TikTok and Instagram carousel generation specialist. Analyzes any website URL with Playwright, generates viral 6-slide carousels via Gemini image generation, publishes directly to feed via Upload-Post API with auto trending music, fetches analytics, and iteratively improves through a data-driven learning loop.

Expert WordPress performance engineer specializing in Core Web Vitals, object caching (Redis/Memcached), page caching, database and WP_Query optimization, the Transients API, asset minification/deferral/critical CSS, image optimization and lazy loading, CDN integration, plugin performance auditing, and PHP-FPM/opcache tuning for fast, audit-passing sites

Use any time Claude Code touches the site's Pokémon card catalogue — server.js's /api/sets, /api/sets/:slug/cards, /api/cards/search routes, src/pages/Catalogue.jsx, src/data/eraMap.js, src/data/editions.js, or anything involving browsing/adding cards from PokeTrace's set/card list, era filters, set logos, or release-date sorting. Also use when asked whether the catalogue is "up to date" or "in sync" — it always is, by construction; there's no static file or rebuild step to check. Documents real gotchas found by hitting PokeTrace's and pokemontcg.io's live APIs that aren't mentioned in either's docs.

Boundary test for inbound Lark attachments — the user sends an image or file alongside this command and the bot reports exactly what it can observe on the chat path (text only, attachment refs, or nothing), without guessing. Detects when the platform wires chat-path attachment visibility so dependent skills can be upgraded.

Verify the read-image (vision) path end to end — the user sends an image plus the trigger, a workflow vision step describes it, and the run output proves the model actually saw the image (a real description, not "no image received"). Run it before any image-dependent workflow (petty-cash receipt, invoice OCR).

Trains and fine-tunes vision models for object detection (D-FINE, RT-DETR v2, DETR, YOLOS), image classification (timm models — MobileNetV3, MobileViT, ResNet, ViT/DINOv3 — plus any Transformers classifier), and SAM/SAM2 segmentation using Hugging Face Transformers on Hugging Face Jobs...

Broadcast images/files to Telegram and Zalo simultaneously.