Installable skillApache-2.0

Communication Draft

Draft personalized follow-up communications (email, call script, SMS) based on account context and history.

BY David-Li0406UPDATED Oct 27, 2026
INTENDED FOR◈OpenClawInstall in a listed agent.

instructions.md

---
name: communication-draft
description: Draft personalized follow-up communications (email, call script, SMS) based on account context and history.
---

# Communication Draft

## When to use this skill

- After identifying account needs follow-up
- User requests a draft for specific account
- Preparing batch follow-up communications

## Channels

### Email
- Subject line + body
- Professional but warm tone
- Include specific context from history

### Call Script
- Talking points, not word-for-word script
- Key questions to ask
- Objection handling notes

### SMS
- Brief, conversational
- Clear call-to-action
- Under 160 characters ideal

## Context to include

1. **From state.md**:
   - Stage and insurance types
   - Next steps and pending actions
   - Contact name

2. **From history.md**:
   - Last interaction summary
   - Previous commitments made

3. **From sources/**:
   - Recent communication topics
   - Client preferences/concerns

## Draft output

```json
{
  "channel": "email",
  "subject": "Following up on your GL quote - Sunny Days Childcare",
  "body": "Hi [Name],\n\nI wanted to follow up on the General Liability quote we discussed last week...",
  "context_used": ["state.md", "call_150734/summary.md"],
  "rationale": "Quote was pitched 4 days ago, client mentioned needing time to review with partner"
}
```

## Templates

See `assets/templates.md` for starter templates by channel and stage.

## Personalization rules

1. Reference specific details from their communications
2. Acknowledge their situation/concerns
3. Provide clear next step
4. Match tone to previous interactions
COMPATIBILITY & REFERENCES

Platform guidance

Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.