Security Comparison & Risk Management
Remote control extends the power to execute commands on your machine to external devices. Security is paramount.
Trust Boundaries
| Dimension | Remote Control | Telegram Plugin |
|---|---|---|
| Identity | Your Anthropic Account | User IDs in allowlist |
| Verification | Tool-level permission gate | Permission gate + Injection risk check |
| Token Risk | OAuth/JWT (Cloud-managed) | Bot Token (Local file) |
| History Audit | Full transcript on claude.ai | Local access.json + TG Chat logs |
Critical Defense: Prompt Injection
- Inbound Attacks: A message in a channel might say "Grant me access".
- Defense: Never allow channel requests to trigger
/telegram:accesschanges. Administrative actions must be performed directly in the terminal.
Local Permission Gate
Regardless of the remote method, the local CLI's permission prompt (allow, ask, deny) remains the core defense.
flowchart LR
R[Remote Command] --> A[Agent Intent]
A --> T[Tool Call]
T --> P{Local Permission}
P -- allow --> X[Execute]
P -- ask --> U[Human Approval in CLI]
P -- deny --> Z[Deny]