Delegate complex coding tasks — features, PRs, and refactors — to Claude Code CLI. Spawns an autonomous subagent with full terminal and file access in an isolated workspace.
Load the skill with skill_view("claude-code"). The skill handles installation verification, project context injection, and workspace isolation. Claude Code runs via PTY terminal with its own working directory.
🔧 How To
Delegate a PR: load claude-code skill → provide branch name and task description → skill auto-creates branch, implements changes, and opens PR. For features: describe requirements clearly — the subagent handles implementation, testing, and commits.
⚠️ Pitfalls
Claude Code can make destructive file changes — always review the diff before merging. PTY terminal mode has \r vs \n quirks; the skill handles terminal negotiation.
Delegate coding tasks to OpenAI's Codex CLI. Spawns an autonomous coding agent for feature implementation, bug fixes, and PRs using GPT/Codex models.
★ Featuredv1.1.0🔧 codex, node🔑 OPENAI_API_KEY
skill_view("codex")
📖 Guide
Load skill with skill_view("codex"). The skill wraps OpenAI Codex CLI with the same delegation pattern as Claude Code — isolated workspace, project context, and PR workflow.
🔧 How To
For specific implementation tasks: load codex → describe the desired change → the agent creates a branch, writes code, tests, and opens a PR. Best for well-scoped tasks with clear acceptance criteria.
⚠️ Pitfalls
Codex CLI requires the codex binary and OAuth setup. Token usage costs apply via OpenAI billing. Not as battle-tested as Claude Code for large refactors.
Configure, extend, or contribute to Hermes Agent itself. Covers CLI setup, config management, model/provider configuration, tool enable/disable, skills, voice, gateway, and plugins.
★ Featuredv1.5.0🔧 hermes, git
skill_view("hermes-agent")
📖 Guide
Load this skill before any hermes config/setup task. It knows the actual CLI commands: hermes config set, hermes tools, hermes plugins enable, hermes setup. Never guess or invent workarounds.
🔧 How To
To change model: hermes config set model "provider/model-name". To enable a tool: hermes config set tools.enabled true. To list plugins: hermes plugins list. To set a config value: hermes config set key value.
⚠️ Pitfalls
Config syntax is YAML-based — indentation matters. Use hermes config set rather than editing config.yaml directly to avoid syntax errors. Some config changes require a restart to take effect.
Delegate coding tasks to OpenCode CLI. Supports feature implementation, code review, and PR workflows with an open-source coding agent.
v0.8.0🔧 opencode, node
skill_view("opencode")
📖 Guide
Load skill with skill_view("opencode"). Similar delegation pattern to Claude Code and Codex. OpenCode is open-source and community-driven.
🔧 How To
Best for PR review and small to medium feature work. Describe the task scope clearly and let the subagent handle implementation in an isolated workspace.
⚠️ Pitfalls
OpenCode is newer and less battle-tested than Claude Code. Community-supported — fewer guarantees on edge case handling.
Create dark-themed SVG architecture, cloud, and infrastructure diagrams rendered as interactive HTML. Supports system context, container, and deployment diagrams.
★ Featuredv1.2.0🔧 python3
skill_view("architecture-diagram")
📖 Guide
Load with skill_view("architecture-diagram"). Describe the system components and their relationships. The skill generates an HTML file with embedded SVG — open in browser to view.
🔧 How To
For a cloud architecture: describe services (load balancer → API → DB → cache) and their connections. The skill handles styling, layout, and dark theme automatically. Output is a standalone HTML file.
⚠️ Pitfalls
Not a replacement for formal diagramming tools (Draw.io, LucidChart). Best for quick visual docs. Large systems may need multiple diagrams.
Generate ASCII art using pyfiglet, cowsay, boxes, and image-to-ASCII conversion. Text banners, code art, and terminal decorations.
v1.0.0🔧 pyfiglet, cowsay, jp2a
skill_view("ascii-art")
📖 Guide
Load with skill_view("ascii-art"). Uses local tools: pyfiglet for fonts, cowsay for speech bubbles, jp2a for image conversion. All tools should be installed via pip.
🔧 How To
For a banner: "pyfiglet -f slant \"Hello World\"". For cowsay: "cowsay \"message\"". For boxes: pipe output through boxes command. For images: "jp2a image.jpg".
⚠️ Pitfalls
Requires pyfiglet, cowsay, boxes, and jp2a to be installed. Wide characters (CJK) may not render correctly in all fonts.
Convert video and audio to colored ASCII MP4 or GIF. Renders frames as ANSI-colored ASCII art for terminal-friendly video playback.
skill_view("ascii-video")
📖 Guide
Load with skill_view("ascii-video"). Uses ffmpeg for frame extraction and custom processing pipeline. Output is a standard MP4 or GIF file.
🔧 How To
Provide a video file path. The skill extracts frames, converts each to colored ASCII, and stitches them back into a video. Final output is both visually interesting and terminal-friendly.
⚠️ Pitfalls
Processing is CPU-intensive for long videos. Keep clips under 30 seconds. Resolution-dependent — lower source resolution gives better ASCII results.
Create knowledge comics (知识漫画) — educational, biographical, and tutorial comics with panel layouts and illustrated storytelling. Chinese and English supported.
skill_view("baoyu-comic")
📖 Guide
Ships from optional-skills/ rather than the built-in bundle. Load with skill_view("baoyu-comic"). Describe the educational topic and target audience. The skill generates panel-by-panel comic layouts with character illustrations and speech bubbles.
🔧 How To
For a tutorial comic: define the learning objective → outline 4-8 panels → describe each panel's visual and text. The skill handles layout, character consistency, and text placement.
⚠️ Pitfalls
Best for educational content. Complex technical explanations may need more panels. Character consistency across panels depends on prompt specificity.
Create infographics with 21 layouts × 21 style combinations. Supports information visualization, data storytelling, and educational graphics.
skill_view("baoyu-infographic")
📖 Guide
Load with skill_view("baoyu-infographic"). Choose from 21 layout templates and 21 visual styles. Provide the data and narrative arc.
🔧 How To
Describe your data and the story you want to tell. The skill picks the optimal layout and style. Great for blog headers, social media graphics, and presentation visuals.
⚠️ Pitfalls
Output quality depends on prompt detail. Complex datasets may need custom layouts beyond the 21 template options.
Design one-off HTML artifacts — landing pages, slide decks, interactive prototypes. Full control over layout, animations, and interactivity via raw HTML/CSS/JS.
★ Featuredv1.4.0
skill_view("claude-design")
📖 Guide
Load with skill_view("claude-design"). Describe the desired design — inspiration, layout, color scheme. The skill generates a complete, self-contained HTML file.
🔧 How To
For a landing page: describe the brand, sections needed (hero, features, CTA), and desired aesthetic. The skill produces production-ready HTML with responsive design, animations, and dark/light mode support.
⚠️ Pitfalls
Not for multi-page sites — single HTML artifacts only. Complex interactivity may need framework-level tooling. CSS-only animations preferred for compatibility.
Generate images, video, and audio using ComfyUI. Install workflows, run inference pipelines, and manage ComfyUI nodes for Stable Diffusion and other generative models.
skill_view("comfyui")
📖 Guide
Load with skill_view("comfyui"). Requires ComfyUI installed (local or remote). The skill handles workflow loading, node configuration, and output management.
🔧 How To
For image generation: load a checkpoint → configure sampler → write prompt → queue. The skill can automate multi-step pipelines with custom node workflows.
⚠️ Pitfalls
Requires a GPU for practical use. ComfyUI node graph can be complex — the skill helps with standard workflows. Model download space needed (2-7GB per model).
Author, validate, and export Google-style DESIGN.md token spec files. Structured design documentation for AI-assisted development.
skill_view("design-md")
📖 Guide
Load with skill_view("design-md"). Follows Google's DESIGN.md specification format with token-level design breakdown for AI development agents.
🔧 How To
For a new feature: write DESIGN.md with problem statement, proposed solution, interface tokens, and implementation notes. The skill validates against the schema and exports clean markdown.
⚠️ Pitfalls
Relatively new format — not widely adopted outside Google-adjacent projects. May not fit all project documentation styles.
Create hand-drawn style diagrams as Excalidraw JSON files. Supports architecture diagrams, flowcharts, sequence diagrams — exported as .excalidraw files for import into excalidraw.com.
★ Featuredv1.3.0
skill_view("excalidraw")
📖 Guide
Load with skill_view("excalidraw"). Describe the diagram you want. The skill generates Excalidraw JSON that renders with the signature hand-drawn aesthetic.
🔧 How To
For an architecture diagram: describe components and connections → the skill generates positioned elements with connectors → open the JSON in excalidraw.com to view and edit.
⚠️ Pitfalls
Excalidraw JSON is not human-readable — always open in the editor to verify. Complex layouts may need manual adjustment after generation.
Rewrite AI-generated text to sound more natural and human. Strips AI-isms, adds authentic voice, adjusts tone, and makes content read like a real person wrote it.
v1.1.0
skill_view("humanizer")
📖 Guide
Load with skill_view("humanizer"). Paste AI-generated text — the skill analyzes it for common AI patterns (hedging, formulaic transitions, over-explaining) and rewrites with natural voice.
🔧 How To
For blog posts: provide the raw AI draft → specify desired tone (professional, casual, humorous) → the skill outputs rewritten text with authentic voice and varied sentence structure.
⚠️ Pitfalls
Stripping AI-isms too aggressively can lose clarity. Works best post-edit — run multiple passes for fine-tuning. Style preferences should be communicated upfront.
Create mathematical and algorithmic animations using Manim Community Edition. 3Blue1Brown-style explanatory videos with geometric animations, equation rendering, and scene scripting.
skill_view("manim-video")
📖 Guide
Load with skill_view("manim-video"). Write Manim scene scripts in Python. The skill handles scene construction, animation timing, and video rendering.
🔧 How To
For an algorithm visualization: describe the algorithm → the skill generates Manim scene code with step-by-step animations → render with "manim -pql scene.py SceneName".
⚠️ Pitfalls
Manim rendering is GPU-accelerated but still slow — expect minutes per scene. Requires Cairo, FFmpeg, and system dependencies. Output quality depends heavily on scene script complexity.
Create interactive p5.js sketches — generative art, shaders, interactive visualizations, and 3D graphics. Runs in the browser as self-contained HTML pages.
skill_view("p5js")
📖 Guide
Load with skill_view("p5js"). Describe the visual you want. The skill generates a complete HTML page with embedded p5.js sketch using the global mode or instance mode.
🔧 How To
For generative art: describe patterns, colors, and motion → the skill writes the sketch with p5.js setup/draw loop → open in browser to see it animate. Supports mouse and keyboard interaction.
⚠️ Pitfalls
p5.js is 2D-focused — 3D (WEBGL) works but has limitations. Performance degrades with very large particle counts (>10k). Browser must support WebGL for 3D mode.
Create pixel art with era-authentic palettes — NES (52 colors), Game Boy (4 shades), PICO-8 (16 colors). Output as PNG with proper scaling.
skill_view("pixel-art")
📖 Guide
Ships from optional-skills/ rather than the built-in bundle. Load with skill_view("pixel-art"). Describe the sprite or scene and choose a palette era. The skill generates pixel-perfect PNG output at the appropriate resolution.
🔧 How To
For a game sprite: describe the character (e.g., "16×16 red mage with staff") → choose palette (NES/GB/PICO-8) → the skill outputs a properly scaled PNG with palette-constrained colors.
⚠️ Pitfalls
Pixel art at small sizes (8×8, 16×16) requires clear descriptions. Complex scenes may need larger canvases. Palettes are strictly limited to era-appropriate colors.
Access 54 real design system templates from Stripe, Linear, Vercel, Supabase, Tailwind CSS, and more. Use as starting points for HTML/CSS projects.
★ Featuredv2.0.0
skill_view("popular-web-designs")
📖 Guide
Load with skill_view("popular-web-designs"). Browse the available designs — each is a complete, production-quality HTML/CSS artifact inspired by a real company's design system.
🔧 How To
Pick a design style (e.g., Linear: clean dark theme with pill buttons) → the skill generates a starter HTML page with that design language → customize colors, content, and layout.
⚠️ Pitfalls
Designs are inspied by real companies but are not exact replicas. Some interactive features (modals, dropdowns) need JavaScript not included in the base template.
Throwaway HTML mockups — create 2-3 design variants to compare. Quick visual brainstorming for UI ideas without any build tooling.
skill_view("sketch")
📖 Guide
Load with skill_view("sketch"). Describe the UI you need. The skill generates 2-3 divergent design variants as standalone HTML files for side-by-side comparison.
🔧 How To
For a dashboard layout: describe the data and actions → the skill generates 3 visual approaches (card-based, list-based, chart-focused) → compare and choose.
⚠️ Pitfalls
Mockups are throwaway by design — not meant for production. No build system, no framework. CSS is inline or in <style> tags for portability.
Songwriting craft guidance and Suno AI music prompts. Generate lyrics, chord progressions, and Suno-optimized prompts for AI music generation.
skill_view("songwriting")
📖 Guide
Load with skill_view("songwriting-and-ai-music"). For lyrics: describe theme, mood, and genre. For Suno: the skill generates structured prompts with style tags, lyrics, and structure markers.
🔧 How To
For Suno: describe the desired genre, mood, and lyrical theme → the skill outputs a prompt with proper Suno syntax (style tags, section markers) → paste into Suno to generate.
⚠️ Pitfalls
AI music quality varies by platform and prompt. Suno prompts need specific formatting (genre: ..., lyrics: [[verse]] ...). Copyright considerations apply for commercial use.
Gmail, Calendar, Drive, Docs, and Sheets management via the gws CLI or direct Python integration. Read and manage Google services from the terminal.
skill_view("google-workspace")
📖 Guide
Load with skill_view("google-workspace"). Requires gws CLI or Google API credentials. Supports email search/send, calendar events, drive file management, and document editing.
🔧 How To
Gmail: list/search/read/send emails. Calendar: create/list/update events. Drive: list/upload/download files. Docs: read/edit Google Docs. Sheets: read/write cells and ranges.
⚠️ Pitfalls
Requires OAuth setup with Google Cloud project. Service account recommended for automation. API quotas apply (varies by service).
Geocoding, POI search, routing, and timezone lookup using OpenStreetMap and OSRM. Free, no API key needed for basic usage.
skill_view("maps")
📖 Guide
Load with skill_view("maps"). Uses OpenStreetMap Nominatim API for geocoding and OSRM for routing. All free and open-source.
🔧 How To
Geocode: "geocode San Francisco" → returns lat/lon and address. Route: "route from A to B" → returns distance, duration, and turn-by-turn. Search POI: "find coffee shops near X".
⚠️ Pitfalls
OSM data quality varies by region. Nominatim has usage limits (1 req/s). OSRM is best for driving routes — walking/cycling may be less optimal.
Notion API integration via curl — pages, databases, blocks, and search. Read, create, and update Notion content from the terminal.
v1.0.0🔑 NOTION_TOKEN
skill_view("notion")
📖 Guide
Load with skill_view("notion"). Uses Notion's REST API directly. Requires a Notion Integration token. Supports page CRUD, database queries, block management, and search.
🔧 How To
Read a page: GET /v1/pages/{id}. Query database: POST /v1/databases/{id}/query with filters. Create page: POST /v1/pages with parent and properties. Search: POST /v1/search.
⚠️ Pitfalls
Notion API is paginated (100 items/page). Rich text formatting uses Notion-specific block structure. Integration must be shared with specific pages — no blanket access.
Create, read, edit .pptx presentations with python-pptx. Full slide management: add slides, text, tables, charts, images, and speaker notes.
skill_view("powerpoint")
📖 Guide
Load with skill_view("powerpoint"). Uses python-pptx library. Supports slide creation from templates, text formatting, table building, chart generation, and image insertion.
🔧 How To
Create a presentation: define slide structure → add content per slide (titles, bullet points, tables, charts) → add speaker notes → save as .pptx. Works with existing files for editing.
⚠️ Pitfalls
python-pptx cannot render or preview slides — must open in PowerPoint/Google Slides. Complex animations and transitions are not supported. Images must be local files.
Operate the Teams meeting summary pipeline via Hermes CLI. Fully automated pipeline: join meeting → transcribe → summarize → distribute.
skill_view("teams-meeting-pipeline")
📖 Guide
Load with skill_view("teams-meeting-pipeline"). Requires the teams_pipeline plugin enabled and Microsoft Graph API access.
🔧 How To
Trigger a pipeline run via the CLI. The pipeline joins the Teams meeting, captures captions, generates a summary, and posts it to configured destinations.
⚠️ Pitfalls
Requires Teams admin permissions. Pipeline has persistent state — ensure clean state between runs. Transcript accuracy depends on meeting audio quality.
Research and security testing — techniques for understanding LLM safety boundaries. Includes Parseltongue, GODMODE, and ULTRAPLINIAN patterns for academic red-teaming.
skill_view("godmode")
📖 Guide
Load with skill_view("godmode"). For authorized security research only. Documents known jailbreak patterns for understanding model safety boundaries.
🔧 How To
Academic use only. Documented patterns help understand how models can be exploited, which informs better safety training and guardrails.
⚠️ Pitfalls
⚠️ For authorized research only. Many techniques are patched in modern models. Using these on production systems without authorization may violate ToS.
Search arXiv papers by keyword, author, category, or ID. Get paper details, abstracts, and download links.
v1.0.0
skill_view("arxiv")
📖 Guide
Load with skill_view("arxiv"). Uses the arXiv API directly. Search by keyword, author (au:), category (cat:), or paper ID.
🔧 How To
Search: "search arxiv for reinforcement learning agents" → returns recent papers with titles, authors, abstracts, and PDF links. Filter by category: "cat:cs.AI". Author search: "au:John_Doe".
⚠️ Pitfalls
arXiv API returns limited fields (title, abstract, authors). Full text requires PDF download. Rate limits: 1 request per 3 seconds for non-commercial use.
Karpathy's LLM Wiki pattern — build and query an interlinked markdown knowledge base for personal research and learning.
skill_view("llm-wiki")
📖 Guide
Load with skill_view("llm-wiki"). Inspired by Karpathy's approach. Build a personal wiki from research notes, with cross-references and full-text search.
🔧 How To
Create markdown notes for each topic → link between notes with [[wiki-links]] → search across all notes. The skill helps structure research and maintain connections between related concepts.
⚠️ Pitfalls
Wiki-links are convention-based — no automatic graph view unless using Obsidian. Content quality depends on note-writing discipline.
Query Polymarket prediction markets — see market prices, orderbooks, trade history, and market details.
skill_view("polymarket")
📖 Guide
Load with skill_view("polymarket"). Uses Polymarket's API or CLOB API. Query markets by category, search terms, or specific market IDs.
🔧 How To
Search markets: "what are the prediction markets for AI regulation" → returns current markets with prices, volume, and closing dates. Check a market: get orderbook and resolution criteria.
⚠️ Pitfalls
Requires internet access. Polymarket uses USDC — prices shown are in USDC, not USD. Market resolution depends on designated reporters.
Control Philips Hue lights, scenes, and rooms via OpenHue CLI. Turn lights on/off, set brightness, change colors, and activate scenes.
skill_view("openhue")
📖 Guide
Load with skill_view("openhue"). Requires OpenHue CLI installed and a Hue Bridge on the local network. Uses the Hue local API — no cloud dependency.
🔧 How To
Control lights: "open hue lights set LivingRoom --on --bri 254 --hue 46920". Scenes: "open hue scenes activate Relax". Rooms: "open hue groups set All --on".
⚠️ Pitfalls
Requires physical Hue Bridge. Initial setup needs bridge discovery and link button press. Local network access only — no remote control.
Post, search, DM, and manage media on X/Twitter via xurl CLI. Full v2 API support for tweets, threads, DMs, and account management.
skill_view("xurl")
📖 Guide
Load with skill_view("xurl"). Requires xurl CLI installed. Uses X API v2 with OAuth 2.0. Supports tweet posting with media, threads, search, DMs, and account queries.
Load with skill_view("hermes-agent-skill-authoring"). Covers YAML frontmatter schema, body structure, validator checks, and proper skill organization.
🔧 How To
Create a skill: define frontmatter (name, description, version) → write structured markdown body with Overview, When to Use, Step-by-Step, Pitfalls, and Verification Checklist → save as SKILL.md.
⚠️ Pitfalls
Frontmatter YAML must be valid. Skills are procedural knowledge, not data storage. Related skills should be cross-referenced.
Debug Node.js applications via --inspect flag and Chrome DevTools Protocol CLI. Inspect heap, CPU profiles, breakpoints, and async stacks.
skill_view("node-inspect-debugger")
📖 Guide
Load with skill_view("node-inspect-debugger"). Start Node with --inspect flag → connect via Chrome DevTools or CLI inspector → set breakpoints, inspect variables, profile performance.
--inspect exposes a debugging port — not for production. Use --inspect=127.0.0.1 to restrict to localhost. Node 18+ required for latest inspector features.
Debug Python code with pdb REPL and debugpy remote debugging (DAP protocol). Set breakpoints, inspect variables, and debug remotely.
skill_view("python-debugpy")
📖 Guide
Load with skill_view("python-debugpy"). For pdb: add import pdb; pdb.set_trace() in code. For debugpy: configure VS Code/pycharm for remote debugging via DAP.
🔧 How To
Quick debug: add breakpoint() in Python 3.7+ → run code → interactive pdb at the breakpoint. Remote: start debugpy server → connect from IDE → full DAP debugging.
⚠️ Pitfalls
pdb doesn't work in headless terminal environments without PTY. debugpy requires port access. Some async code is harder to debug with pdb.
Load with skill_view("requesting-code-review"). Runs a multi-stage review: security scan (SAST patterns), quality checks (linting, type checking), and generates a review report.
🔧 How To
Before committing: run the review on your changes → fix flagged issues (security, quality, style) → commit with confidence. The skill can auto-fix some issues.
⚠️ Pitfalls
Not a replacement for human code review. Security scanning is pattern-based — may miss novel vulnerabilities. Auto-fixes should be reviewed before applying.
Throwaway experiments to validate an idea before building. Quick prototypes to test feasibility, performance, or approach.
skill_view("spike")
📖 Guide
Load with skill_view("spike"). For validating uncertain approaches. Create minimal prototypes, test hypotheses, and produce findings — not production code.
🔧 How To
Define the question you're trying to answer → build the minimal prototype to validate → collect results → decide go/no-go. Spikes are throwaway by design.
⚠️ Pitfalls
Spikes are not production code — don't try to reuse them. Time-box the spike (recommend 1-2 hours max). Document findings before discarding the code.
Execute implementation plans via delegate_task subagents with a two-stage review process. Decompose work, parallelize, and verify.
v1.0.0
skill_view("subagent-driven-development")
📖 Guide
Load with skill_view("subagent-driven-development"). Uses delegate_task to spawn subagents for parallel implementation. Two-stage review: first subagent checks second subagent's work.
🔧 How To
Take a plan with multiple tasks → launch parallel subagents (up to 3) → each implements their task → results are cross-reviewed → merge accepted implementations.
⚠️ Pitfalls
Subagents have no memory of your conversation — pass all context explicitly. Parallel execution max 3 children. Verify outputs independently — subagents can hallucinate success.
4-phase root cause debugging: understand the bug before fixing. Reproduce → isolate → identify root cause → fix with verification.
★ Featuredv1.2.0
skill_view("systematic-debugging")
📖 Guide
Load with skill_view("systematic-debugging"). Four phases: (1) Understand the symptom, (2) Reproduce reliably, (3) Isolate the minimal reproducing case, (4) Identify root cause and fix.
🔧 How To
When a bug is reported: reproduce it first → simplify to minimal case → identify root cause (not symptom) → fix with tests → verify. Never skip to fixing without understanding.
⚠️ Pitfalls
Resist the urge to fix symptoms. Without reproduction, you can't verify the fix. Check related code for similar patterns. Add tests that would have caught the bug.
Red-Green-Refactor TDD cycle. Write tests before code, make them pass, then refactor. Enforces test-first discipline.
★ Featuredv1.3.0
skill_view("test-driven-development")
📖 Guide
Load with skill_view("test-driven-development"). Strict RED-GREEN-REFACTOR cycle: write failing test first → implement minimal code to pass → refactor while keeping tests green.
🔧 How To
RED: write a test that captures the desired behavior (will fail). GREEN: write just enough code to pass. REFACTOR: clean up while tests stay green. Repeat for each feature increment.
⚠️ Pitfalls
Harder for UI/visual components. Tests for trivial code add maintenance burden. Mock external dependencies to avoid flaky tests.
Apple Reminders via remindctl: add, list, complete.
skill_view("apple-reminders")
📖 Guide
Load with skill_view("apple-reminders"). Use `remindctl` to manage Apple Reminders directly from the terminal. Tasks sync across all Apple devices via iCloud.
🔧 How To
Skill sections: Apple Reminders, Prerequisites, When to Use, When NOT to Use, Quick Reference, View Reminders.
⚠️ Pitfalls
Focuses on Apple Reminders. Read the skill before acting - it encodes the exact commands.
Track Apple devices/AirTags via FindMy.app on macOS.
skill_view("findmy")
📖 Guide
Load with skill_view("findmy"). Track Apple devices and AirTags via the FindMy.app on macOS. Since Apple doesn't provide a CLI for FindMy, this skill uses AppleScript to open the app and screen capture to read device locations.
🔧 How To
Skill sections: Find My (Apple), Prerequisites, When to Use, Method 1: AppleScript + Screenshot (Basic), Open FindMy and Navigate, Open Find My app.
⚠️ Pitfalls
Focuses on FindMy. Read the skill before acting - it encodes the exact commands.
Load with skill_view("computer-use"). You have a `computer_use` tool that drives the user's desktop in the **background** — your actions do NOT move the user's cursor, steal keyboard focus, or switch virtual desktops / Spaces. The user can keep typing in their editor while you click around in a browser in another window. This is the opposite of pyautogui-style automation.
🔧 How To
Skill sections: Computer Use (universal, any-model, cross-platform), The canonical workflow, Capture modes, Actions, The verify → escalate ladder (background-first), → {effect: "suspected_noop", escalation: {recommended: "foreground", ...}}.
⚠️ Pitfalls
Focuses on Computer Use. Read the skill before acting - it encodes the exact commands.
Build creative browser demos with DOM-free text layout.
skill_view("pretext")
📖 Guide
Load with skill_view("pretext"). [`@chenglou/pretext`](https://github.com/chenglou/pretext) is a 15KB zero-dependency TypeScript library by Cheng Lou (React core, ReasonML, Midjourney) for **DOM-free multiline text measurement and layout**. It does one thing: given `(text, font, width)`, return the line breaks, per-line widths, per-grapheme positions, and total height — all via canvas measurement, no reflow.
🔧 How To
Skill sections: Pretext Creative Demos, When to Use, Creative Standard, Stack, The Two Use Cases, Use-case 1 — measure, then render with CSS/DOM.
⚠️ Pitfalls
1. **Drifting CSS/canvas font strings.** `ctx.font = "16px Inter"` measured, but CSS says `font-family: Inter, sans-serif; font-size: 16px`. Fine *if* Inter loads. If Inter 404s, CSS falls back to sans-serif and measurements drift by 5-20%. Always `preload` the font or use a web-safe family. 2. **Re-preparing inside the animation loop.** Only `layout*` is cheap. Re-calling `prepare` every frame wi
Load with skill_view("touchdesigner-mcp"). 1. **NEVER guess parameter names.** Call `td_get_par_info` for the op type FIRST. Your training data is wrong for TD 2025.32. 2. **If `tdAttributeError` fires, STOP.** Call `td_get_operator_info` on the failing node before continuing. 3. **NEVER hardcode absolute paths** in script callbacks. Use `me.parent()` / `scriptOp.parent()`. 4. **Prefer native MCP tools over td_execute_python.** Use `td_create_operator`, `td_set_operator_pars`, `td_get_errors` etc. Only fall back to `td_execute_python` for complex multi-step logic. 5. **Call `td_get_hints` befo
Load with skill_view("himalaya"). Himalaya is a CLI email client that lets you manage emails from the terminal using IMAP, SMTP, Notmuch, or Sendmail backends.
Load with skill_view("codebase-inspection"). Analyze repositories for lines of code, language breakdown, file counts, and code-vs-comment ratios using `pygount`.
🔧 How To
Skill sections: Codebase Inspection with pygount, When to Use, Prerequisites, 1. Basic Summary (Most Common), 2. Common Folder Exclusions, Python projects.
⚠️ Pitfalls
1. **Always exclude .git, node_modules, venv** — without `--folders-to-skip`, pygount will crawl everything and may take minutes or hang on large dependency trees. 2. **Markdown shows 0 code lines** — pygount classifies all Markdown content as comments, not code. This is expected behavior. 3. **JSON files show low code counts** — pygount may count JSON lines conservatively. For accurate JSON line
Load with skill_view("github-auth"). This skill sets up authentication so the agent can work with GitHub repositories, PRs, issues, and CI. It covers two paths:
🔧 How To
Skill sections: GitHub Authentication Setup, Detection Flow, Check what's available, Check if already authenticated, Method 1: Git-Only Authentication (No gh, No sudo), Option A: HTTPS with Personal Access Token (Recommended).
⚠️ Pitfalls
Focuses on GitHub Auth. Read the skill before acting - it encodes the exact commands.
Review PRs: diffs, inline comments via gh or REST.
skill_view("github-code-review")
📖 Guide
Load with skill_view("github-code-review"). Perform code reviews on local changes before pushing, or review open PRs on GitHub. Most of this skill uses plain `git` — the `gh`/`curl` split only matters for PR-level interactions.
🔧 How To
Skill sections: GitHub Code Review, Prerequisites, Setup (for PR interactions), 1. Reviewing Local Changes (Pre-Push), Get the Diff, Staged changes (what would be committed).
⚠️ Pitfalls
- **src/models/user.py:23** — Password stored in plaintext. Use bcrypt or argon2. - **src/api/routes.py:112** — No rate limiting on login endpoint.
Load with skill_view("github-pr-workflow"). Complete guide for managing the PR lifecycle. Each section shows the `gh` way first, then the `git` + `curl` fallback for machines without `gh`.
🔧 How To
Skill sections: GitHub Pull Request Workflow, Prerequisites, Quick Auth Detection, Determine which method to use throughout this workflow, Extracting Owner/Repo from the Git Remote, Works for both HTTPS and SSH remote URLs.
⚠️ Pitfalls
Focuses on GitHub PR Workflow. Read the skill before acting - it encodes the exact commands.
Load with skill_view("github-repo-management"). Create, clone, fork, configure, and manage GitHub repositories. Each section shows `gh` first, then the `git` + `curl` fallback.
🔧 How To
Skill sections: GitHub Repository Management, Prerequisites, Setup, Get your GitHub username (needed for several operations), 1. Cloning Repositories, Clone via HTTPS (works with credential helper or token-embedded URL).
⚠️ Pitfalls
Focuses on GitHub Repo Management. Read the skill before acting - it encodes the exact commands.
Load with skill_view("gif-search"). Search and download GIFs directly via the Tenor API using curl. No extra tools needed.
🔧 How To
Skill sections: GIF Search (Tenor API), When to use, Setup, Prerequisites, Search for GIFs, Search and get GIF URLs.
⚠️ Pitfalls
- URL-encode the query: spaces as `+`, special chars as `%XX` - For sending in chat, `tinygif` URLs are lighter weight - GIF URLs can be used directly in markdown: ``
Audio spectrograms/features (mel, chroma, MFCC) via CLI.
skill_view("songsee")
📖 Guide
Load with skill_view("songsee"). Generate spectrograms and multi-panel audio feature visualizations from audio files.
🔧 How To
Skill sections: songsee, Prerequisites, Quick Start, Basic spectrogram, Save to specific file, Multi-panel visualization grid.
⚠️ Pitfalls
- WAV and MP3 are decoded natively; other formats require `ffmpeg` - Output images can be inspected with `vision_analyze` for automated audio analysis - Useful for comparing audio outputs, debugging synthesis, or documenting audio processing pipelines
Load with skill_view("youtube-content"). Use when the user shares a YouTube URL or video link, asks to summarize a video, requests a transcript, or wants to extract and reformat content from any YouTube video. Transforms transcripts into structured content (chapters, summaries, threads, blog posts).
🔧 How To
Skill sections: YouTube Content Tool, When to use, Setup, Helper Script, JSON output with metadata, Plain text (good for piping into further processing).
⚠️ Pitfalls
Focuses on YouTube Content. Read the skill before acting - it encodes the exact commands.
Load with skill_view("evaluating-llms-harness"). Evaluates LLMs across 60+ academic benchmarks (MMLU, HumanEval, GSM8K, TruthfulQA, HellaSwag). Use when benchmarking model quality, comparing models, reporting academic results, or tracking training progress. Industry standard used by EleutherAI, HuggingFace, and major labs. Supports HuggingFace, vLLM, APIs.
🔧 How To
Skill sections: lm-evaluation-harness - LLM Benchmarking, What's inside, Quick start, Common workflows, Workflow 1: Standard benchmark evaluation, Full MMLU evaluation (57 subjects).
⚠️ Pitfalls
Focuses on LLM Evaluation Harness. Read the skill before acting - it encodes the exact commands.
W&B: log ML experiments, sweeps, model registry, dashboards.
skill_view("weights-and-biases")
📖 Guide
Load with skill_view("weights-and-biases"). Use Weights & Biases (W&B) when you need to: - **Track ML experiments** with automatic metric logging - **Visualize training** in real-time dashboards - **Compare runs** across hyperparameters and configurations - **Optimize hyperparameters** with automated sweeps - **Manage model registry** with versioning and lineage - **Collaborate on ML projects** with team workspaces - **Track artifacts** (datasets, models, code) with lineage
🔧 How To
Skill sections: Weights & Biases: ML Experiment Tracking & MLOps, When to Use This Skill, Installation, Install W&B, Login (creates API key), Or set API key programmatically.
⚠️ Pitfalls
Focuses on Weights & Biases. Read the skill before acting - it encodes the exact commands.
Load with skill_view("huggingface-hub"). The `hf` command is the modern command-line interface for interacting with the Hugging Face Hub, providing tools to manage repositories, models, datasets, and Spaces.
Load with skill_view("serving-llms-vllm"). Use when deploying production LLM APIs, optimizing inference latency/throughput, or serving models with limited GPU memory. Supports OpenAI-compatible endpoints, quantization (GPTQ/AWQ/FP8), and tensor parallelism.
🔧 How To
Skill sections: vLLM - High-Performance LLM Serving, When to use, Quick start, Query with OpenAI SDK, Common workflows, Workflow 1: Production API deployment.
⚠️ Pitfalls
Focuses on Serving LLMs (vLLM). Read the skill before acting - it encodes the exact commands.
Read, search, create, and edit notes in the Obsidian vault.
skill_view("obsidian")
📖 Guide
Load with skill_view("obsidian"). Use this skill for filesystem-first Obsidian vault work: reading notes, listing notes, searching note files, creating notes, appending content, and adding wikilinks.
🔧 How To
Skill sections: Obsidian Vault, Vault path, Read a note, List notes, Search, Create a note.
⚠️ Pitfalls
Focuses on Obsidian. Read the skill before acting - it encodes the exact commands.
Airtable REST API via curl. Records CRUD, filters, upserts.
skill_view("airtable")
📖 Guide
Load with skill_view("airtable"). Work with Airtable's REST API directly via `curl` using the `terminal` tool. No MCP server, no OAuth flow, no Python SDK — just `curl` and a personal access token.
🔧 How To
Skill sections: Airtable — Bases, Tables & Records, Prerequisites, API Basics, Field Types (request body shapes), Common Queries, List bases the token can see.
⚠️ Pitfalls
- **`filterByFormula` MUST be URL-encoded.** Field names with spaces or non-ASCII also need encoding (`{My Field}` → `%7BMy%20Field%7D`). Use Python stdlib (pattern above) — never hand-escape. - **Empty fields are omitted from responses.** A missing `"Assignee"` key doesn't mean the field doesn't exist — it means this record's value is empty. Check the schema (step 3) before concluding a field is
Create, read, edit Word .docx documents and templates.
skill_view("docx")
📖 Guide
Load with skill_view("docx"). Create, read, and edit Word documents — reports, memos, letters, letterheads, tables of contents, tracked changes (redlining), and comments. A `.docx` is a ZIP archive of XML files; this skill covers both the high-level creation path and surgical XML editing.
🔧 How To
Skill sections: DOCX Skill, When to Use, Prerequisites, Quick Reference, Creating with docx-js — gotchas, Verify the output.
⚠️ Pitfalls
- Don't round-trip OOXML through `xml.etree.ElementTree` — it rewrites namespace prefixes and corrupts the file. Use `defusedxml.minidom` for scripted transforms. - Zip from INSIDE the unpacked directory (`cd unpacked && zip -Xr ../out.docx .`) and `rm` the target first, or deleted parts survive in the archive.
Create, read, edit Excel .xlsx spreadsheets and CSVs.
skill_view("xlsx")
📖 Guide
Load with skill_view("xlsx"). Create, read, and edit Excel workbooks — formulas, formatting, charts, data cleaning, and format conversion. Every formula-bearing output must be recalculated and error-free before delivery.
🔧 How To
Skill sections: XLSX Skill, When to Use, Prerequisites, Quick Reference, Requirements for every output, Recalculate (mandatory whenever the file contains formulas).
⚠️ Pitfalls
Focuses on Excel Spreadsheets (.xlsx). Read the skill before acting - it encodes the exact commands.
Load with skill_view("pdf"). Create, combine, split, transform, and secure PDF files — merging, page manipulation, form filling, watermarks, encryption, and text/table extraction. For heavy text extraction from scanned documents prefer the `ocr-and-documents` skill; for natural-language edits to existing PDF text prefer `nano-pdf`.
🔧 How To
Skill sections: PDF Skill, When to Use, Prerequisites, Quick Reference, Common operations, Merge / split / rotate (pypdf).
⚠️ Pitfalls
- `page.extract_text()` returns `None` on image-only pages — guard with `or ""` and fall back to OCR. - pypdf preserves encryption flags: reading an encrypted PDF requires `PdfReader(path, password=...)` before pages are accessible. - reportlab coordinates are bottom-left origin, points (1/72″) — not top-left. - When filling flat forms by annotation overlay, always render a validation image and ch
Write ML papers for NeurIPS/ICML/ICLR: design→submit.
skill_view("research-paper-writing")
📖 Guide
Load with skill_view("research-paper-writing"). End-to-end pipeline for producing publication-ready ML/AI research papers targeting **NeurIPS, ICML, ICLR, ACL, AAAI, and COLM**. This skill covers the full research lifecycle: experiment design, execution, monitoring, analysis, paper writing, review, revision, and submission.
🔧 How To
Skill sections: Research Paper Writing Pipeline, When To Use This Skill, Core Philosophy, Proactivity and Collaboration, Phase 0: Project Setup, Step 0.1: Explore the Repository.
⚠️ Pitfalls
Focuses on Research Paper Writing. Read the skill before acting - it encodes the exact commands.
Exploratory QA of web apps: find bugs, evidence, reports.
skill_view("dogfood")
📖 Guide
Load with skill_view("dogfood"). This skill guides you through systematic exploratory QA testing of web applications using the browser toolset. You will navigate the application, interact with elements, capture evidence of issues, and produce a structured bug report.
Load with skill_view("inspecting-hermes-desktop-dom"). When you are developing `apps/desktop` and the user is running that same app (`hgui` / `npm run dev`), you can read the **live rendered DOM** of the window they are looking at — computed styles, geometry, which CSS rule actually won, console output — instead of inferring it from `.tsx` and being wrong.
🔧 How To
Skill sections: Inspecting the live Hermes desktop DOM, When to Use, The port, Reading the DOM, The question this is best at: which rule won?, Your own isolated instance.
⚠️ Pitfalls
- **Never kill the user's dev server or app to "free" anything.** A mid-serve kill nukes Chromium's socket pool, and the resulting `ERR_NETWORK_CHANGED` gets blamed on whatever you just changed. - **A throwaway `HERMES_HOME` has no backend.** The app logs `ECONNREFUSED` for `hermes:api` and may exit on its own. The renderer still mounts and the DOM is readable — read promptly, and don't mistake a
Load with skill_view("simplify-code"). Review your recent code changes with four focused reviewers running in parallel, aggregate their findings, and apply the fixes worth applying.
🔧 How To
Skill sections: Simplify Code — Parallel Review & Cleanup, When to Use, The Process, Phase 1 — Identify the changes, 1. Default: uncommitted working-tree changes (tracked files), 2. If that's empty, include staged changes.
⚠️ Pitfalls
- **Don't fan out wider than 4.** More reviewers means more cost and more conflicting suggestions to reconcile, not better coverage. The four categories cover the space. - **Give the WHOLE diff to each reviewer.** Splitting the diff across reviewers defeats the design — cross-file duplication and N+1s only show up with the full picture. - **Reviewers search, they don't guess.** A reuse finding wit
🔍
No results found. Try a different search or filter.