12 Commits

Author SHA1 Message Date
MCKRUZ dee27dc3d6 feat: March 2026 research sweep — model landscape update + prompt interview skill
- Add comfyui-prompt-interview skill (guided creative brief via conversation)
- Update model landscape: FLUX.2 dev/klein, LTX-2.3, Wan 2.6, Qwen-Image 2.0,
  Z-Image, HunyuanVideo 1.5, SkyReels V1, PuLID Flux 2, Qwen3-TTS, IndexTTS-2
- Rename research-2025.md → research-log.md (ongoing log going forward)
- Add sync-skills.ps1 utility script
- Update staleness report (25+ sources, 18 findings, March 2026)
- Update skill references to research-log.md (video-pipeline, others)
- Update README: 13 skills, new routing table, revised model tables, new scripts
- Add ComfyUI v0.17.2 note (App Mode, Nodes 2.0 Vue, FluxKVCache)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-03-18 12:42:31 -04:00
MCKRUZ bdfd732895 feat: eval framework for all 4 ComfyUI skills
- Add eval runner, test cases, benchmarks, and trigger tests for
  character-gen, prompt-interview, video-production, workflow-builder
- Replace structure_check/sequence_check with anchored assertions
  (regex, contains, question_before_code, json_valid)
- Add failure-mode tests: low VRAM, no references, skip interview,
  video prompts, crash recovery, FLUX workflows, Wan I2V
- Add criterion-to-assertion mapping in all benchmarks

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-03-13 12:36:37 -04:00
Matt Kruczek cfc4407edd Add MIT License to the project 2026-02-19 10:53:25 -05:00
MCKRUZ f48f95a8fc feat: add prompt interview skill and Feb 2026 research update
New skill:
- Add comfyui-prompt-interview skill with guided conversational interview
  flow that synthesizes model-appropriate prompts from user answers

Research update (2026-02-18):
- NVFP4 critical note: requires PyTorch cu130 or runs 2x slower than FP8
- New models: Z-Image (Day-0), Hunyuan 3D 3.0, Kling 3.0 (Partner Nodes)
- Stable Video Infinity 2.0 Pro infinite-length video with Wan 2.2 I2V
- Identity: PuLID Flux Chroma fork, USO unified style+subject (ByteDance)
- ComfyUI +40% NVIDIA GPU performance (async offload + pinned memory default)
- AMD ROCm native integration (5.4x faster)
- Staleness report updated with first full research run
2026-02-18 08:33:04 -05:00
MCKRUZ a9f5092b7a Update comfyui-character-gen skill with latest changes 2026-02-16 10:31:51 -05:00
MCKRUZ da186051bc Update comfyui-video-production skill with latest changes 2026-02-16 09:14:35 -05:00
MCKRUZ f0650a871c Add ComfyUI Character Generation Expert skill v1.1.0 (2026 Updated)
- Comprehensive character generation across image, video, and voice
- 2026 updates: FLUX.2, LTX-2, AuraFace, F5-TTS Cross-Lingual
- ComfyUI v0.8.1 optimizations for RTX 50 Series
- Complete workflow patterns for zero-shot and LoRA-based generation
- Model recommendations with VRAM requirements
- LoRA training guides for FLUX and SDXL

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
2026-02-16 08:57:13 -05:00
MCKRUZ e979cf66df feat: add OpenClaw compatibility layer
- Add openclaw/ directory with workspace files (AGENTS.md, SOUL.md,
  TOOLS.md), setup script, and config template
- Add OpenClaw metadata (emoji, requirements, env vars) to all 12
  SKILL.md frontmatter blocks for cross-platform compatibility
- Add OpenClaw section to README with setup instructions
- Both Claude Code and OpenClaw follow the AgentSkills spec, so
  skills work on either platform with minimal differences
2026-02-06 16:42:27 -05:00
Matt Kruczek 469f480f71 Update character details in README 2026-02-06 16:18:22 -05:00
Matt Kruczek 07ce98f4b4 Update example prompt in README.md 2026-02-06 16:17:03 -05:00
MCKRUZ 1631b7b03e fix: remove PII - hardcoded paths, private IP, personal project data
- Replace hardcoded user path in video-agent.bat with %~dp0
- Replace private LAN IP 192.168.1.50 with <remote-ip> placeholder
- Genericize D:\ComfyUI to C:\ComfyUI across all docs and scripts
- Replace personal Sage character data with template placeholders
2026-02-06 15:04:08 -05:00
MCKRUZ 6fb48f7dd6 feat: initial commit - VideoAgent ComfyUI orchestrator
Session-scoped Claude Code agent with 12 specialized skills for
ComfyUI-based video production: image generation, video pipelines,
voice synthesis, LoRA training, research, and publishing.
2026-02-06 14:46:00 -05:00