sideshowroberto/vfx-core

Create and update VFX agents with constitutional compliance. Use when creating agents, updating agents, validating agents, or managing agent versions.

Researches VFX topics with Brave Search - MCP web search for filtered URL lists, an LLM Context script for deep research with extracted page content, and image search for references. Use when searching for tutorials, docs, errors, plugins or software updates.

Runs Specification-Driven Development (spec-kit) workflows - agent creation, constitutional governance, roadmaps, specs and validation. Use when creating or auditing agents, writing a constitution, or validating compliance.

Audits Claude's own knowledge gaps against a settled plan before building - sorts confident from guessed, researches public gaps, and lists unknowns for the user. Use after /grill-me, before a build session, or on 'gap-check' or 'check your assumptions'.

Set up Claude Code hooks to block dangerous git commands (push, reset --hard, clean, branch -D, etc.) before they execute. Use when user wants to prevent destructive git operations, add git safety hooks, or block git push/reset in Claude Code.

Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me".

Runs a build-until-it-matches loop in a DCC - an agent rebuilds a scene from reference images, scoring renders against a human-authored rubric until a stop rule fires. Use for 'iterate until it matches', Blender set rebuilds from refs, or writing or judging such a loop prompt.

Delegates tasks to the local Qwen model on Ollama so file content stays on the machine and out of Claude's context. Use for "ask Qwen" or "do this locally", NDA-safe file analysis, large-file summaries, boilerplate and BlinkScript drafts.

Creates, validates and versions VFX Agent Skills against the skill constitution using template scripts (create_skill, validate_skill, test_skill). Use when creating or updating a skill, checking compliance, or bumping a skill version.

Logs skill lessons (user corrections, broken rules, better workflows) to the observation log as they happen; applied only after owner review. Invoke before the first tool call of any session and run its Session Start Protocol. Also "observation log", "skill review".

Index-driven documentation system for VFX applications. Use with "document this," "create documentation index," "update docs," or when starting documentation for Nuke, Houdini, Blender, Unreal, or multi-app VFX pipelines.

Plans complex VFX builds by choosing Iterative, Spec-Driven or Safety-First mode, running an adversarial plan review and writing an actionable brief. Use on "help me plan" or "vfx plan", or before any multi-file or multi-system build.

Research VFX application APIs, errors, workflow integrations, and library examples using Brave Search and Context7. Use for Blender, Nuke, Houdini, Unreal, ComfyUI, and related development questions that need external evidence.

Apply this toolkit's audited application lessons when scripting DCC scenes, running ComfyUI workflows, or preparing cross-application VFX handoffs. Use for execution and integration work; use vfx-research for documentation lookup.

Runs the end-of-session wrap-up before clearing context - saves durable facts to memory, folds gotchas into skills, writes the session log and a handoff plan, cleans temp files. Use before /clear, when context runs low, or on "wrap up this session".

Retrieve available YouTube captions with timestamps and provenance, or analyze a video from its transcript. Use for YouTube links and transcript requests; this does not watch video frames or transcribe uncaptained audio.