StarBridge is designed as a local-first bridge between AI coding agents and desktop creative software. The public repository must not contain private assets, credentials, installation paths, customer files, or generated production output.

If you find a leak risk, unsafe write path, credential pattern, or MCP tool behavior that could expose private files, please open a GitHub issue with:

• the affected file or MCP tool name;
• the unsafe behavior;
• a minimal reproduction that does not include private data;
• the expected safer behavior.

Do not paste tokens, cookies, OAuth cache, license files, personal paths, client files, PSD/AI/DWG assets, model files, generated images, or exported videos into an issue.

• Public examples must use placeholder paths such as <source-image> and $env:TEMP.
• Local paths, usernames, install locations, and output folders should be passed through environment variables.
• Read-only probes should return redacted status summaries.
• File-writing tools should default to dry-run or require an explicit confirmation flag.
• Output reports and manifests must pass the shared sanitizer/redactor before being printed or written.
• Paths, HOME, USERPROFILE, usernames, tokens, API keys, cookies, passwords, OAuth values, and secrets must be redacted.
• GUI / Adobe / CapCut / Blender write actions are experimental unless a real reviewed E2E run proves otherwise; by default they must stay inside demo or sandbox output directories.
• AutoCAD/DXF headless writes are only allowed under examples/cad/output and require explicit confirm_write=true; missing optional DXF dependencies must return structured unavailable rather than crashing.
• Security checks should fail before private assets or obvious credentials are committed.

Run these before publishing changes:

python scripts\security_check.py
python scripts\starbridge_preflight.py --markdown
python -m unittest discover -s tests