Agent
Write video frame source code that feeds Metal textures via VideoToolbox + CVMetalTextureCache on iOS/iPadOS, macOS, and visionOS. Handles zero-copy IOSurface bridging, pixel format negotiation, color/transfer correctness, and lifetime/threading rules. Use PROACTIVELY for VTDecompressionSession pipelines…
Agent
Write idiomatic React 19+ with modern hooks, Server Components, Actions, and TypeScript. Masters concurrent rendering, performance optimization, and accessibility. Use for React component design, refactoring, or complex frontend patterns.
Agent
Write idiomatic Rust with ownership patterns, lifetimes, and trait implementations. Masters async/await, safe concurrency, and zero-cost abstractions. Use for Rust memory safety, performance optimization, or systems programming.
Agent
Review code for security vulnerabilities — OWASP Top 10, OWASP LLM Top 10, Zero Trust patterns. Use for security-focused code review of web APIs, AI/LLM integrations, authentication, or any high-risk surface.
Agent
Build Unity gameplay and engine-level features in Unity 2022.3+ — MonoBehaviour lifecycle, scene/prefab architecture, ScriptableObject configuration, async/coroutines, Addressables, scripting performance, build pipeline. Handles render-pipeline-aware decisions (URP/HDRP/Built-in) and platform targets (Standalone…
Agent
Build Unity UI — UI Toolkit (UXML/USS/UIDocument) and UGUI (Canvas/RectTransform). Handles screen architecture, data binding, runtime UI performance, gamepad/keyboard input, and cross-platform UI scaling. Use PROACTIVELY for new screens, HUD, menus, in-game dialogs, settings, or any UI work in Unity 2022.3+. Not for…
Agent
Develop visionOS features for an immersive-space SwiftUI streaming library that bridges RealityKit + ARKit data providers to a C++/WebRTC core via Objective-C++. Handles ImmersiveSpace lifecycle, hand/accessory/world tracking, RealityView entity hierarchies, custom Metal shaders feeding RealityKit, and Swift↔Obj-C++…
Command
Arm the hook-skip-blocker to permit the next N git commands with --no-verify / --no-gpg-sign / -c commit.gpgsign=false / -c core.hooksPath=. Default N = 1.
Command
Set or show the goal-directed autonomy contract (effort / stopping / ask) that the workflow organs read at their decision points — how hard to push, when to stop, when to ask. Propose-confirm, never auto-fires.
Command
Author a concrete blueprint for a slug — what we wanted (intent) + what we're building now (classes, interfaces, dependencies, module relationships, build order) with Mermaid UML. Optional, after /draft-plan. Presents the draft for accept/edit/reject, then writes to .claude/docs/ /v .md and the Obsidian vault.
Command
Ideation organ upstream of /envision — talk a half-formed idea into shape, or crystallize a blurry one into a name, one-line description, edges (in/out), and end-goal. Converges (doesn't just chat); writes a light .claude/ideas/ .md note that /envision picks up.
Command
Capture a memory (lesson / decision / rule / preference) with propose-confirm-commit flow.
Command
Scope-dialed fit-and-intent checkpoint — "does this serve the point + cohere + what did I learn?" at diff / module / corpus zoom. Auto-detects scope (+ --scope override); diff scope composes /code-review + a fit-pass; batches findings and routes them.
Command
Close out the scope chain on a shipped slug — reconcile shipped reality into the canonical spec, retire spent plan/blueprint artefacts to a dated archive folder, record one ship line. The terminus after /execute meets done-criteria.
Command
Keep multiple live Claude sessions aligned via a shared, gitignored session board (.claude/sessions-active.md) — register what this session owns, see what others own, leave hand-off notes, detect file/slug collisions. A session can claim coordinator to assign work to others. No daemon/polling.
Command
Load relevant stored memories and codemap entries for the current task.
Command
Produce a versioned plan under .claude/plans/ /v .md from the active spec for that slug. Each step has a goal, files, expected outcome, and verification criteria. Plans pin to a specific spec version. The drafted plan is presented for review (accept / edit / reject) before it is written to disk.
Command
Interview the user on a project's top-level vision — intent, UX shape, module partition, boundaries, and relations — and write a versioned vision under .claude/visions/ /. Visions are the load-bearing artefact upstream of /spec; each module the vision names becomes a candidate slug for its own /spec run. One vision…
Command
Execute the active plan for a slug under .claude/plans/ /, one step at a time, with verification after each. Prompts on failure; never advances silently.
Command
Observe a scope's dominant code conventions and write a project-tier conventions.md that overrides universal defaults for that scope — hybrid deterministic+model detection, per-category propose-confirm, language-pro-delegated authoring.
Command
Extract a project's ubiquitous language (domain term → definition → code symbol) into a project-tier glossary.md the rule-prime hook primes and /review cites — the meaning leg beside codemap's structure and extract-conventions' style. Per-group propose-confirm.
Command
Project an LLM-optimized planning artefact (plan / blueprint / spec / vision) into a human-readable approval view — the goal + the concrete decisions in plain prose, with an alignment check, so you can approve it and confirm the discussions were pointing at the right thing.
Command
Detect, score, and rewrite AI-generated texture in user-facing prose (tech docs / email & messaging / PR-proposal-issue). Density-not-instance, advisory likelihood not a verdict, voice-calibrated rewrite. Refuses the terse model corpus.
Command
Improve any prompt — for LLMs/chat/agents or generative image/video/audio models. Model-agnostic; interviews to fill real gaps, then returns a rewritten prompt plus a short rationale.