draft-a-reply

Draft a Reply

When to use

• You say "draft a reply for {conversation id}" or "draft my response."
• check-my-inbox surfaced thread in morning brief, you clicked in.
• Triaged conversation has status open / waiting_founder, no draft.md yet.
• Never called to send - this skill drafts only.

Connections I need

I run external work through Composio. Before this skill runs I check that the categories below are linked. Missing → I name the category, ask you to connect it from the Integrations tab, stop.

• Inbox (Gmail / Outlook) - pull the live thread and sample your sent replies for tone. Required.
• CRM (HubSpot / Attio / Salesforce) - pull plan, owner, account record for the dossier I read before drafting. Optional if customers.json is already populated.
• Billing (Stripe) - pull monthly revenue for billing-flavored replies. Optional.

If your inbox isn't connected I stop and ask you to connect Gmail or Outlook first.

Information I need

I read your support context first. For every required field that's missing I ask ONE plain-language question (best modality: connected app > file drop > URL > paste) and wait.

• Support context doc - Required. Why I need it: product surface, routing rules, response-time tiers, forbidden phrases all live there. If missing I ask: "Want me to walk you through setting up your support context first? It's a quick interview."
• Voice samples - Required. Why I need it: drafts in the wrong voice get rewritten anyway. If missing I ask: "Want me to pull 10 to 20 of your recent replies to learn your tone, or can you paste 3 to 5 examples?"
• The thread itself - Required. Why I need it: I draft against the actual customer message, not a paraphrase. If missing I ask: "Which conversation should I draft for - share the customer's name or the latest email?"

Steps

1. Read context/support-context.md. If missing/empty, stop and tell me run set-up-my-support-info first.
2. Load thread from conversations/{id}/thread.json. Identify latest customer message - draft respond to that.
3. Chain look-up-a-customer view=dossier for customer on thread. Pull: plan, monthly revenue, open bug-candidates, open followups (from followups.json), any churn-flags.json entry, last 3 conversations from history.
4. Sample voice. Read config/voice.md. If missing or sampleCount < 5, run calibrate-my-voice first. Mirror tone cues: greeting, sign-off, sentence length, whether I use customer's first name. No "I apologize for the inconvenience." No corporate hedging.
5. Draft reply. Match ask:
   • Bug - acknowledge, confirm repro if possible, state next step. Never promise fix date I haven't approved - say "I'll get back to you with a timeline."
   • How-to - answer crisp, link KB article if one exists at articles/{slug}.md (check before linking).
   • Billing - state facts, propose action (refund / credit / plan change). Escalate to me before committing.
   • Churn language - tight, honest, no guilt. Offer genuine option; never promise what not policy in context/support-context.md.
6. Append dossier snippet to conversations/{id}/notes.md (plan, monthly revenue, open bugs, churn status) so I have context when approving.
7. Write conversations/{id}/draft.md atomically. Update conversations.json entry: status = waiting_founder, refresh updatedAt.
8. Append to outputs.json with type: "reply-draft", domain: "inbox", title = "Reply to {customer} re {subject}", summary = opening line, path.
9. Chain track-my-promises. If draft contains commitment ("I'll check with engineering by Friday", "I'll ship next week"), run track-my-promises so due date lands in followups.json.