Labels inbound replies by intent (positive, objection, not-now, out-of-office, referred contact, unsubscribe, auto-ack) and drafts the correct follow-up per label. Works with any provider; gated API only upgrades the label confidence, never the output you see.
Installs into .claude/skills of the current project.
Are you the author of Reply Intent Classifier?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/veneris-io-reply-intent-classifier)
---
name: reply-intent-classifier
description: Labels inbound replies by intent (positive, objection, not-now, out-of-office, referred contact, unsubscribe, auto-ack) and drafts the correct follow-up per label. Works with any provider; gated API only upgrades the label confidence, never the output you see.
---
# Reply Intent Classifier
Classifies reply text on a strict intent rubric, then drafts the follow-up that treats the prospect's signal honestly — never with canned "circling back" filler.
## Labels
- `positive` — explicit interest / tells you to proceed or meet
- `objection` — pushback on fit, timing, cost, need, or authority
- `not-now` — deprioritized without hostility
- `out-of-office` — away mode with a date or implied return
- `referred-contact` — asks you to speak to someone else or offers a referral
- `unsubscribe` — wants out, formally or informally
- `auto-ack` — automated acknowledgment only; no human intent present
## Process
1. Classify to a label with a confidence level (rubric-based; optionally upgraded via gated classifier endpoint if the user has a key)
2. Draft per label:
- **positive** → move straight to a concrete next step (offer 2 slots or a question that shortens the path to a meeting), under 60 words
- **objection** → acknowledge their point in their nouns and address it once, then stop; no counter-argument stacking
- **not-now** → respect and land a concrete re-entry ("I'll come back in Q4 when [consequence]"); do not continue a sequence
- **out-of-office** → pause; nothing sent until their return
- **referred-contact** → thank them, verify the new contact's persona fit before outreach, report back to the original thread
- **unsubscribe** → don't sequence, don't argue; respect and log
- **auto-ack** → only reassess if human intent appears elsewhere in the thread
3. Drafts read as the user's voice. Never draft a bot-style reply pretending it isn't reviewed.
## Approval
Return: **label, confidence, one-line rationale, and the drafted response**. Wait for approval before anything is sent.
## Two-way rule
All labels and response text stay local to the user's conversation. The gated API only upgrades label confidence; it does not collect or store reply contents by default.
