If you discover a security issue in this repository (e.g., a snippet that demonstrates an unsafe pattern, a template that ships an over-broad permission set, or a skill that recommends an insecure practice), please report it responsibly.

Report: Open a private security advisory on GitHub.

Please include:

• Description of the vulnerability
• Steps to reproduce
• Which skill, rule, snippet, or template is affected
• Any suggested fix

This repository ships Markdown skill files, MDC rule files, Python snippets, three starter templates (a Blender extension add-on, a headless batch script, and a GLB-in engine-ready asset pipeline), runnable headless examples under examples/, showcase scripts under showcase/, and an opt-in second plugin, blender-live-session (live-session/), that registers the upstream Blender Lab MCP server. The primary security concerns are:

• Snippets or templates demonstrating insecure patterns (executing arbitrary code from .blend files, loading remote scripts without validation, leaking filesystem paths into logs).
• The extension-addon template declaring over-broad permissions in blender_manifest.toml (e.g. network, files, clipboard, camera) without a documented justification.
• Skills recommending insecure practices (running eval() on driver expressions from untrusted sources, disabling Blender's auto-execute-script protection in headless workflows, embedding credentials in .blend custom properties).
• Headless batch scripts that pass user-controlled input to subprocess or os.system without sanitization.
• Examples and showcase scripts that do any of the above; they run headless in CI and users run them locally.
• The blender-live-session plugin: its pin and launch command in live-session/mcp.json, and its skill's guidance. That server executes agent-generated Python in a running Blender without guards. See live-session/SECURITY.md for what is in scope there; flaws in the upstream server itself belong upstream.

Issues with the Blender Python API itself (bpy, bmesh, bpy_extras) belong upstream at https://projects.blender.org and are out of scope here.

This is a content repository — Markdown, MDC, Python, and TOML files consumed directly by an AI agent. The core plugin installs no runtime and starts no server (the opt-in blender-live-session plugin launches the upstream MCP server from a pinned commit; this repository does not host or patch it), and there is no long-lived deployment to patch, so there are no maintenance branches and no backports.

A confirmed fix lands on main and ships in the next release cut by release.yml. Most consumers install the plugin, which comes from the plugin-dist branch: release.yml force-pushes that branch as a single orphan commit on each release that changes plugin content, and the branch carries no tags, so a plugin install cannot be pinned to an older release. The remedy is to update the plugin (Claude Code: /plugin marketplace update blender-developer-tools; Cursor local plugin: git fetch origin plugin-dist then git reset --hard FETCH_HEAD in its folder, since a force-pushed orphan commit cannot be pulled). A checkout of main pins by release tag (vX.Y.Z) or tracks main, and its remedy is to move to the current release. Nothing older is patched in place.

We aim to acknowledge reports within 48 hours and provide a fix or mitigation within 7 days for confirmed issues.