v0.5.0 - Re-pin skills to ultralytics v8.4.174 and audit every claim (#26)

🌟 Summary

v0.5.0 refreshes the Ultralytics YOLO agent skills for ultralytics 8.4.174, with clearer guidance on YOLO26 inference, model export, training, and Platform workflows. This is a skills/plugin documentation update—not a release of new model weights or runtime code.

📊 Key Changes

• YOLO26 inference and export guidance updated: The skills replace the deprecated end2end setting with nms. By default, YOLO26 uses its one-to-many head with NMS; nms=False selects the NMS-free head, while nms=True embeds NMS in supported exports. The docs explain how this affects iou, output formats, and older end2end arguments.
• Export options and compatibility clarified: The format guide adds AMD Xilinx export support and updates Apple Core AI requirements and format-specific NMS behavior.
• Model and training references refreshed: The catalog adds guidance for SAM 3.1 and other model assets, while training notes cover options such as quantization-aware training and updated validation behavior.
• Platform instructions corrected and expanded: The skills clarify dataset histogram bins and unlabeled-image counts, improve dataset search and image-similarity guidance, and document important workflow details such as rename effects and upload limits.
• Plugin and CI maintenance: The plugin version is bumped to 0.5.0, with a fresh Codex ZIP and SBOM planned for release; CI now relies on the setup-uv action’s defaults.

🎯 Purpose & Impact

• Helps users choose the right YOLO26 inference mode and avoid confusion around the deprecated end2end setting.
• Makes model export and deployment guidance more accurate across supported runtimes and hardware.
• Reduces the risk of errors when inspecting datasets, managing Platform resources, or following training instructions.
• Gives AI agents and users documentation checked against the pinned Ultralytics package and current Platform references, making the skills more reliable.

What's Changed

• Use setup-uv defaults for Python and the venv by @glenn-jocher in #25
• Clarify dataset histogram bin ranges by @JaviChulvi in #21
• Re-pin skills to ultralytics v8.4.174 and audit every claim by @glenn-jocher in #26

Full Changelog: v0.4.0...v0.5.0

v0.4.0 - Clear directory findings, polish skill copy, and release v0.4.0 (#24)

🌟 Summary

v0.4.0 makes the Ultralytics YOLO agent-skills pack easier to validate, install, and use across coding agents—with clearer Platform CLI guidance and end-to-end YOLO workflows.

📊 Key Changes

• 🔐 Prioritized in PR #24: Replaced the curl-to-Python command used to inspect the public Platform API schema with guidance to fetch only the relevant schema using the agent’s available tools. This avoids directory scanners’ security warnings while preserving the CLI guidance.
• 📦 Added versioned plugin packaging for Claude Code and Codex, plus marketplace listings. The release workflow now creates a Codex upload ZIP and an SPDX software bill of materials.
• 💬 Updated Codex starter prompts to guide users from trying YOLO, to training a model, to deploying or exporting it.
• 🧰 Expanded the skills pack to cover the full lifecycle: model selection, datasets, training and tuning, inference and tracking, export, and Platform CLI workflows—including managing Platform resources and Agents.
• 🤖 Added practical guidance for both Ultralytics Platform and local Python/CLI workflows. Model catalogs recommend YOLO26 as the default while also covering other model families and deployment formats.

🎯 Purpose & Impact

• ✅ Users can install the skills through supported agent plugin marketplaces and get clearer starting points for common tasks.
• 🛡️ Agents can consult Platform API details without relying on a command pattern that triggered directory security checks.
• 🚀 The skills provide more complete, practical guidance for moving from dataset preparation to training and deployment.
• ℹ️ This is a skills and packaging release, not a change to YOLO model architectures or weights.

What's Changed

• Add Ultralytics YOLO agent skills by @JaviChulvi in #2
• Canonicalize Ultralytics URLs by @glenn-jocher in #3
• Tailor repository setup for public YOLO skills by @glenn-jocher in #4
• Expand YOLO skills with Platform workflows by @glenn-jocher in #5
• Ground skill catalogs against Ultralytics 8.4.119 by @glenn-jocher in #7
• Correct YOLO skill defaults and backend guidance by @JaviChulvi in #6
• Add skills-specific PR template and refresh bug report placeholder by @glenn-jocher in #8
• Remove trailing slashes and update redirected Ultralytics URLs by @pderrenger in #9
• Accept normalized CLA signature comments by @glenn-jocher in #10
• Update skills guidance for Ultralytics 8.4.138 by @JaviChulvi in #11
• Add platform-cli skill for the ul CLI by @JaviChulvi in #12
• Include platform-cli in issue templates and focus agent guidance by @glenn-jocher in #13
• Expand Platform CLI inspection guidance by @JaviChulvi in #14
• Document the ul cloud workflow shortcuts by @glenn-jocher in #15
• Update yolo cloud shortcuts for detached training and downloads by @JaviChulvi in #16
• Document platform-cli facts that replace help lookups by @JaviChulvi in #17
• Refresh AGENTS.md skill checklist and version pin locations by @glenn-jocher in #18
• Add Agents commands to the platform-cli skill by @JaviChulvi in #19
• Pass required name when editing Agents in platform-cli by @glenn-jocher in #20
• Fit Codex and Claude directory listing rules: starter prompts, icons, setup-uv CI by @glenn-jocher in #23
• Clear directory findings, polish skill copy, and release v0.4.0 by @glenn-jocher in #24

New Contributors

• @glenn-jocher made their first contribution in #3
• @JaviChulvi made their first contribution in #2
• @pderrenger made their first contribution in #9

Full Changelog: e5e8483...v0.4.0