security-gate

Classifying and gating inbound Telegram messages involves screening content in real-time to determine safety levels. This Skill evaluates incoming prompts and assigns a classification of ALLOW, WARN, or BLOCK with a brief rationale to prevent unsafe messages from reaching Claude Code.

Deterministic message classification applies predefined rules to evaluate Telegram message content without external tool access. It outputs a JSON object containing the classification decision, confidence score, reason, and any flagged patterns, ensuring consistent and auditable safety gatekeeping.

No external tool access is required to screen Telegram prompts for unsafe content. This gatekeeping Skill operates as a lightweight, standalone filter before relaying messages, applying policy-driven rules to proactively reduce the attack surface without external dependencies.

The best way to audit safety decisions for Telegram message screening is to use policy-driven classification with attached rationale. This approach outputs a JSON object with the classification, confidence level, and reason for each message, supporting compliance and auditability.

Use a lightweight gatekeeping filter when you need to proactively screen inbound Telegram messages before they reach Claude Code. It is ideal for reducing the attack surface with deterministic rules and minimal overhead, rather than deploying a complex external security model.