Repository navigation
[Security] Telegram Bridge command injection #118
Description
Telegram Bridge Command Injection
File: scripts/telegram-bridge.js:86-88
Problem: User messages are directly appended to SSH commands.
const escaped = message.replace(/'/g, "'\''");
const cmd = export NVIDIA_API_KEY='${API_KEY}' && nemoclaw-start openclaw agent --agent main --local -m '${escaped}' --session-id 'tg-${sessionId}';
Impact: Attackers can execute commands within the sandbox by sending malicious messages.
API Key may be compromised.
Exploitation Requirements:
Requires obtaining a Telegram Bot Token.
Requires being configured in the whitelist (or not having ALLOWED_CHAT_IDS configured).
Requires knowing the sandbox name.
Remediation Recommendations: Use parameterized messaging. Implement strict whitelist validation for user input.
Activity
security: fix command injection in Telegram bridge (Fixes NVIDIA#118)
security: fix command injection in Telegram bridge (Fixes NVIDIA#118)
im23pds commentedon Mar 17, 2026
thanks Fix
@23pds
SlowMist CISO
security: fix command injection in Telegram bridge (Fixes NVIDIA#118)
ericksoa commentedon Mar 23, 2026
Tracking upstream as openclaw/openclaw#16323, which covers the broader attack surface of untrusted messaging platform users reaching shell execution paths. We're working with the OpenClaw team to address this — closing in favor of tracking there.
test(security): add Brev E2E tests for command injection and credenti…
test(security): add Brev E2E tests for command injection and credenti…
test(security): add Brev E2E tests for command injection and credenti…
test(security): add Brev E2E tests for command injection and credenti…