Agent Skills

A collection of skills for AI coding agents. Skills are packaged instructions that extend agent capabilities with domain-specific knowledge.

Skills follow the Agent Skills format.

Install

npx skills add vladmdgolam/agent-skills

Or install a specific skill:

npx skills add vladmdgolam/agent-skills --skill cinema4d-mcp

Available Skills

🎬 cinema4d-mcp

Cinema 4D MCP expert for extracting scene data, writing C4D Python scripts, and controlling Cinema 4D through MCP tools.

Use when:

• Using Cinema 4D MCP tools (get_scene_info, list_objects, execute_python_script, etc.)
• Writing Python scripts for C4D extraction or manipulation
• Working with MoGraph cloners, effectors, and fields
• Baking animation data from C4D scenes
• Debugging C4D Python API errors

Covers:

• 7 critical C4D API gotchas (world/local coords, visibility constants, sequential stepping, etc.)
• MoGraph extraction and animation track discovery patterns
• When to use structured MCP tools vs execute_python_script
• Known error tables with fixes (Python API + MCP tool errors)
• Redshift availability matrix
• Timeout management and chunked baking strategies

🧊 blender-mcp

Blender MCP expert for scene inspection, Python scripting, GLTF export, and material/animation extraction.

Use when:

• Using Blender MCP tools (get_scene_info, execute_python, screenshot, etc.)
• Writing Blender Python scripts for scene extraction or manipulation
• Exporting scenes to GLTF/GLB for web (Three.js, R3F)
• Debugging material or texture export losses
• Optimizing GLB files with gltf-transform
• Using asset integrations (PolyHaven, Sketchfab, Hyper3D Rodin, Hunyuan3D)

Covers:

• 7 critical rules (export timeouts, modifier handling, Draco pitfalls, name mapping, etc.)
• Scene hierarchy, material, and animation extraction patterns
• Headless CLI export (bypasses MCP timeout)
• Material export survival matrix (what survives GLTF, what doesn't)
• Texture optimization pipeline (resize → WebP → Draco)
• Known error tables with fixes (MCP, export, Python API, texture paths)

⏱️ time-lens

Analyze and visualize time spent on software projects by combining data from multiple sources: WakaTime coding time, git commit session detection, Claude Code usage, Codex CLI usage, and Cursor IDE usage.

Use when:

• Analyzing work hours or calculating time spent on a project
• Generating a work hours report or visualizing coding activity
• Creating a project time breakdown
• Summarizing development effort across date ranges

Covers:

• 5 data sources: WakaTime API, git sessions, Claude Code prompts, Codex CLI prompts, Cursor IDE prompts
• Interactive HTML dashboard (dark-themed, Chart.js)
• Markdown report with ASCII charts
• Reconciliation logic across overlapping sources

🖨️ pdf-look-scanned

Make PDF documents look like they were scanned on a physical scanner, with optional signature replacement.

Use when:

• Making a PDF look scanned (grayscale, noise, blur, rotation, edge shadows)
• Replacing digital signatures with real handwritten ones from a source PDF
• Preparing documents that need a "printed and scanned" appearance

Covers:

• Configurable scan effects (DPI, noise, blur, contrast, JPEG quality)
• Signature extraction from source PDFs (phone photos, scans)
• Multi-signature replacement on different pages
• Coordinate discovery workflow for signature placement
• Metadata considerations

🎨 figma-context-mcp

Expert guide for using the Figma Context MCP (Framelink) efficiently while avoiding 429 rate-limit errors.

Use when:

• Using Framelink Figma MCP tools (get_figma_data, download_figma_images)
• Fetching Figma designs for code generation
• User hits 429 rate limits from Figma API
• Extracting design tokens or component data from Figma
• Downloading Figma assets/images
• Any design-to-code workflow involving Figma URLs

Covers:

• Metadata-first, prune-first, fetch-last pipeline (2-3 API calls instead of dozens)
• Depth-limited node tree fetching to keep responses <500 KB
• Batch image downloads with deduplication
• Plan-tier awareness (limits tied to file owner's plan, not yours)
• Rate-limit diagnostics and recovery steps
• Common anti-patterns that cause 429 lockouts
• Alternative workflows when rate-limited (screenshots, manual export)

📚 apple-books-mcp

Extracts highlights, annotations, and book data from Apple Books via MCP.

Use when:

• Exporting book highlights with colors and notes
• Extracting annotations from specific books or a reading list
• Searching highlights across your entire library
• Matching books from an external list against Apple Books library

Covers:

• Efficient batch color extraction (5 API calls instead of 150+)
• Color map building via get_highlights_by_color + Python cross-referencing
• Annotation data structure (style values, color mapping, underlines)
• Markdown export patterns with color-coded emoji indicators
• Book matching strategies (translations, partial titles, multiple editions)
• Troubleshooting common issues (empty results, null colors, large outputs)

👁️ visual-feedback-loop

AI agent visual feedback loop for capturing and inspecting rendered output from any web app during iterative development.

Use when:

• Iterating on visual code (UI, canvas, 3D, SVG, charts, animations)
• You need to SEE the rendered result to evaluate quality
• Doing visual QA or A/B comparison of render variants

Covers:

• Dev-screenshot API pattern (GET trigger → SSE → client capture → POST back → file saved)
• Parameterized offscreen rendering with custom params
• WebMCP tool registration for Chrome Canary agents
• Console and Chrome MCP fallbacks
• Next.js reference implementation

🪛 reverse-engineer-js

LLM-assisted deobfuscation of minified or obfuscated JavaScript bundles. Wraps humanifyjs with the right defaults, the pseudo-TTY workaround, and patterns for both whole-bundle and single-module workflows.

Use when:

• Deobfuscating bundle.js, fxhash projects, or any minified JS
• The user says "humanify this", "deobfuscate this", "rename mangled vars", "what does bundle.js do"
• Working in a project with an og/ or modules/ folder fed by humanify
• humanify crashed with cursorTo is not a function

Covers:

• Default model: gemini-3.1-flash-lite-preview
• Pseudo-TTY workaround (script -q /dev/null ...) for non-interactive shells
• Single-module extraction from large bundles
• webcrack pre-split for webpacked bundles
• Sourcemap-first check (often beats LLM rename for free)
• Common backends: gemini / openai / local

🔍 agent-sessions

Search, list, and resume AI agent sessions across Claude Code, Codex CLI, Gemini CLI, opencode, Hermes Agent, and Cline CLI from the terminal.

Use when:

• Finding a past conversation ("find that session where I worked on...")
• Reading latest/last messages from a session id (ses_..., UUID, or Cline id)
• Resuming a previous Claude Code session by ID
• Listing recent activity across agents
• Searching session history for a topic or project

Covers:

• CLI tool at ~/Play/radar/tools/agent-sessions with --agent, --search, --search-scope, --project filters
• Direct handoff flow: --pack <session-id> --max-chars 8000 before manual filesystem or database inspection
• Compact JSON output for scripting; --include-search-text is opt-in for transcript indexes
• Compact handoff snippets via --context <session-id...>
• Bounded evidence packs and excerpts via --pack <id...> and --read <id...> --query; use --role user for user-only messages
• Claude session resume via --resume <id>
• Data sources: ~/.claude/projects/, ~/.codex/sessions/, ~/.gemini/tmp/*/chats/, ~/.local/share/opencode/opencode.db, ~/.hermes/state.db, ~/.cline/data/db/sessions.db
• Companion tools: claude-history (TUI), Claude Code History Viewer (GUI)

🌊 fluid-interfaces

Design and review principles for fluid, gestural, physically-responsive interfaces, distilled from Apple's WWDC 2018 "Designing Fluid Interfaces" talk (the team that built the iPhone X gestural UI).

Use when:

• Designing or reviewing touch/gesture interactions (drag, swipe, dismiss, dock)
• Tuning spring-based motion, elastic/rubberband behaviors, or scroll physics
• Something "feels off", "feels janky", or "doesn't feel native"
• Deciding between timed animations and continuous spring behaviors

Covers:

• 8 core principles: instant response, redirectability/interruption, spatial consistency, gesture hinting, lightweight-input/amplified-output, rubberbanding, smooth frames of motion, behavior-over-animation
• Spring tuning via damping/response instead of raw mass/stiffness/duration
• Momentum projection formula for "throw to nearest endpoint" interactions (with Swift code)
• Gesture mechanics checklist: taps, swipes, one-to-one tracking, continuous feedback, resolving competing gestures
• Teaching techniques for gestural UI (visual cues, elevation, paired animations, explanations, designing for play)

⚡ threejs-perf-loading

Performance and loading patterns for real-time Three.js/WebGL/WebGPU sites, based on modern approaches and best practices from top-notch studios and developers (Ivress, Threejspunk, igloo.inc, Noomo Agency) rather than generic advice.

Use when:

• Eliminating the loading-spinner-to-scene freeze/jank (uncompiled shader pipelines hitting the first real frame)
• Designing an adaptive-quality or perf-budget system, or desktop/mobile post-processing tiers
• Tuning scroll/camera damping so input can't "outrun" a max speed
• Deciding where to spend a render-loop's cost (MRT routing, layer-split passes, on-demand shadows, GPU-resident particles, proximity gating)

Covers:

• Shader/pipeline warmup: force every material to compile behind the loading screen before firing the reveal event, including per-camera compileAsync for layer-masked cameras
• Render-loop cost discipline: MRT emissive routing for selective bloom, layer-split passes, shadowMap.autoUpdate = false, GPU-resident particle motion, proximity + hysteresis gating, fused post-processing passes
• Adaptive quality: one flat perf-budget config object, one-way FPS-latched degradation, UA-based gating, genuinely different desktop/mobile post tiers
• Scroll/input smoothing: step-clamped damped lerp (the maxSpeed primitive behind drei's ScrollControls) vs. direct authored-timeline scrubbing
• Transmission/glass materials: owning your render target for a static-fallback trick, and why a shared pre-blurred buffer beats a hand-optimized per-fragment sampler

Adding Skills

Each skill lives in skills/<skill-name>/ with a required SKILL.md and optional references/, scripts/, and assets/ directories.

License

MIT