#102 · RCE / command injection in any service

Any path that lets an attacker execute arbitrary code or shell commands on a worker, container, or host. Highest priority class.

Likely surfaces

• signal-bot: command parsing, file-search → S3 presigned URL generation, integration callbacks (Outline, PeerTube), !ai/!summarize LLM prompts that reach eval-like sinks, signal-cli wrappers, child_process.spawn calls
• task-exchange: webhook → task automation
• search-mcp / bcw-mcp: MCP tool argument parsing
• uxs-portal: MuJoCo physics WS endpoint, FastAPI eval paths
• selfhost services that exec user input

Examples that would qualify

• A bot command argument that reaches child_process.exec unsanitized
• A worker-side template that interpolates user input into eval() or Function()
• A docker-compose or shell script in selfhost/ that does bash -c "$USER_INPUT"
• Container escape via shared volume or sock

--- *This is NOT a normal bounty. Do not post a public PoC or a public PR demonstrating the vulnerability. Submit findings via the bug-report form at https://score.irregulars.io/contribute#report-bug, or email security@irregularchat.com, or DM an admin on Signal. Awards are admin-determined based on impact: low=50 / medium=150 / high=300 / critical=500 / rce=750.*