Only the latest release of Robot Reel receives security fixes. Older tags are not patched; upgrade to the current release before reporting.
Report vulnerabilities privately through GitHub's private vulnerability reporting for this repository:
https://github.com/noteflowai/robot-reel/security/advisories/new
Do not open a public issue for security problems. Include the affected subcommand or page, a reproducible command, and the observed behaviour. There is no dedicated e-mail address; the advisory form is the only private channel.
The robot-reel mcp subcommand runs an MCP director server over stdio. It
exposes filesystem-backed tools (inspect_recording, create_storyboard)
whose source and output paths are meant to stay inside the workspace passed
with --root. scripts/check_director_mcp.py exercises these boundaries in CI.
Any way to read or write outside that workspace root through the MCP tools is
in scope and should be reported.
Also in scope:
- Path handling in the CLI exporters (
direct,compare,blender,newton,vla,stress) that lets a crafted recording or plan write outside the chosen--outputdirectory. - Injection into the exported self-contained HTML pages through crafted recording or storyboard data.
Out of scope:
- Issues in third-party simulators, models, or Blender itself; report those upstream.
- Use of the optional billable Bedrock agent (
--agent) with the reporter's own credentials.