CtrlK
BlogDocsLog inGet started
Tessl Logo

classify-reply

Use when classifying supplied prospect replies for intent, objections, opt-outs, uncertainty, and routing boundaries.

70

Quality

85%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

SecuritybySnyk

Low

Low-risk findings worth noting

SKILL.md
Quality
Evals
Security

Classify Reply

Purpose

Classify what a reply actually says without expanding consent or turning a stop into an objection.

Required inputs

  • Original reply text, author, and date
  • Independently verified contact eligibility
  • Current contact-specific opt-out or stop-contact status
  • Supplied conversation context and routing labels

Mark absent inputs Missing. Preserve the original reply exactly, including punctuation; never reconstruct it.

Workflow

  1. Quote the complete original message verbatim.
  2. Separate explicit facts from interpretation and missing information.
  3. Assign one primary intent: positive interest, information request, meeting request, objection, referral, clear negative, opt-out, or unclear. Add secondary objection tags only when supported.
  4. Give confidence and explain uncertainty. Do not force certainty or infer identity, authority, purchase intent, consent, contactability, or availability.
  5. Route contact-specifically. Return terminal STOP with no copy for an opt-out, stop request, or clear negative. Never downgrade an opt-out to an objection. Do not treat a conditional objection as a clear negative without explicit refusal.
  6. For any potentially reply-eligible route, require verified eligibility and current opt-out status before permitting later drafting. If either is missing, route NEEDS VERIFICATION; if failed, route STOP.
  7. State that classification does not authorize drafting, sending, booking, scheduling, or system updates.

Output format

For each reply return:

  • Original quote: exact text
  • Facts, Inferences, and Missing
  • Primary intent
  • Secondary objection tags
  • Confidence and uncertainty reason
  • Route: STOP, NEEDS VERIFICATION, or bounded human review
  • Copy: None

For every terminal route, name the exact stop reason and prohibit downstream copy or alternate-channel workarounds.

Guardrails

  • Never invent or alter evidence, identity, eligibility, contactability, consent, calendar state, availability, authority, urgency, or citations.
  • Never target from protected traits or unnecessary sensitive data.
  • Never interpret interest, an information request, or conditional language as meeting consent.
  • Never recommend overcoming, delaying, or bypassing a clear negative or opt-out.
  • Never browse, send, book, schedule, suppress records, use credentials, or mutate an external system.

Quality check

Confirm every quote is exact; facts, inferences, and missing information are distinct; primary intent and secondary tags are supported; uncertainty is explicit; opt-outs and clear negatives stop with no copy; ambiguous objections remain distinguishable; and no external action occurred.

Example invocation

Use classify-reply on the supplied Avery, Jordan, and Rowan replies. Preserve each quote and return only evidence-bound classifications and routes.

Repository
llaskin/AI-SDR-Skill-Pack
Last updated
First committed

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.