consensus-send-email-guard

What This Skill Does

consensus-send-email-guard is a sophisticated governance layer designed for AI-driven outbound communication. Its primary function is to intercept email drafts before they reach the SMTP gateway, subjecting them to a multi-agent 'persona panel' evaluation. By aggregating votes based on weighted persona reputation, the skill makes deterministic decisions: APPROVE, BLOCK, or REWRITE.

Beyond simple filtering, the guard acts as an audit engine. It writes decision artifacts and persona reputation updates to the board ledger, ensuring every email sent is traceable. It enforces hard-block policies against sensitive data leakage, legal guarantees, or medical advice—critical issues that often arise from unmonitored LLM outputs. By normalizing the decision-making process, this skill transforms risky raw generation into audited, high-confidence external actions.

Installation

The skill is distributed via the OpenClaw ecosystem and can be integrated into your workspace using the clawhub CLI. Ensure your environment has access to node and tsx for runtime execution:

clawhub install openclaw/skills/skills/kaicianflone/consensus-send-email-guard

Ensure your board state directory is configured in your project settings, as the skill requires write-access to the consensus-state path for artifact persistence.

Use Cases

• Customer Support Automation: Preventing AI agents from promising discounts or refunds that contradict company policy.
• Legal & Compliance Gates: Validating outbound communications in fintech or healthcare to ensure compliance with strict industry regulations.
• Automated Marketing Campaigns: Ensuring brand voice consistency by filtering out hallucinated product features or incorrect price points before mass distribution.

Example Prompts

1. "Evaluate the draft email regarding the quarterly refund request: {email_draft_json}. Apply legal-compliance persona set."
2. "Run the consensus-send-email-guard in external_agent mode using the provided votes from the sentiment analysis and security audit agents."
3. "Review the pending email draft for policy violations using persona_set_id: 'enterprise-security-001'. Log the result to the board."

Tips & Limitations

• Reputation Logic: The skill’s power lies in reputation. Regularly update your persona sets to calibrate which models/agents have authority over specific topics.
• Idempotency: Always rely on the built-in idempotent check to ensure that network blips don’t cause a double-log of decisions to the ledger.
• Latency: Because this performs persona aggregation, expect a slight overhead compared to direct streaming. This is an intentional design trade-off for safety.
• Scope: This tool does not send the email itself; it provides the final decision gate. You must integrate it into your delivery pipeline after generation but before the SMTP relay.