kai-sdr-reply-triage — Reply Loop

Kai root note: knowledge/, harness/, and scripts/ paths in this skill live in the Kai install, not the user's project. Resolve them against the first ancestor directory of this SKILL.md that contains a knowledge/ folder (the Kai plugin root, ~/.claude/kai, or the kai-cmo-harness repo). MARKETING.md, memory/, and any output files live in the current project. If a referenced scripts/ command is not available in this install, say so, skip it, and continue with the file-based guidance — never fabricate its output.

Objective

Every inbound reply ends up as a triage record that names exactly one category, the consent and suppression consequence, the next safe action with an owner, and — only where it is allowed — a drafted response. The record makes the state change explicit: what this reply did to consent, suppression, routing, meeting status, objection state, and CRM state.

The risk this skill exists to remove is a reply that gets answered when it should have been suppressed, escalated, or left alone.

Done when

Work type internal-research — floor E2/C2/O0 (harness/eco-floors.yaml). The triage record itself never leaves the workspace.

• E2 — the reply record exists at its package path and satisfies harness/skill-contracts/sdr-reply-triage.yaml.
• C2 — banned_word_check passes; category, risk tier, suppression action, and evidence fields are all populated.
• O0 — no outcome obligation on the record.

Sending the drafted response is separate work at the cold-email floor E5/C4/O3: provider send receipt plus suppression-list reconciliation, and CAN-SPAM/GDPR/CASL identity, opt-out, and consent basis verified before send.

Constraints

• Missing context means a hold, not a draft. Without sender identity, suppression list, message history, and source evidence for the contact, write a triage hold note instead of a send-ready response.
• Honor opt-outs immediately. Suppress globally; draft nothing beyond an opt-out confirmation where one is required.
• Do not argue with complaints. Stop the sequence, suppress, route to a human owner.
• Do not reply to bounces. Mark the contact path invalid and block follow-up on that address.
• No invented context — no prior relationship, no new claims without evidence.
• Sensitive, legal, or regulated replies do not get routed without human review.
• No CRM mutation. Notes, tasks, and field changes are proposals.
• Under 120 words unless the user asks for a longer sales response. One CTA, one next action.
• Memory candidates carry a Do not promote until condition and stay unpromoted until it is met.

Categories — classify exactly one primary:

Category	Action
interested	Prepare meeting handoff and short response. Trigger /kai-sales-meeting-prep.
objection	Identify objection type and draft one respectful response.
referral	Capture referred person, source, and permission trail.
not_now	Draft low-pressure close and future reminder task.
wrong_person	Ask for routing only when appropriate. Do not pressure.
opt_out	Suppress globally. Do not draft further outreach except opt-out confirmation if required.
bounce	Mark invalid contact path and block follow-up on that address.
complaint	Stop sequence, suppress, and route to human owner.
unsubscribe_confirmed	Record suppression and no further action.
needs_human_review	Use for legal, regulated, sensitive, hostile, ambiguous, or high-value replies.

Context

Need	Load
Record structure, required fields, gate thresholds	harness/skill-contracts/sdr-reply-triage.yaml
Package schema, ledger status values, memory ledger	harness/skill-contracts/sdr-package.yaml
Opt-out, sender identity, consent, suppression law	harness/references/cold-email-rules.md
Account row, contact row, source evidence, message history, suppression state	workspace/sdr-operator/<package-slug>/
Booked meeting or buying conversation follow-on	/kai-sales-meeting-prep

Output goes to workspace/sdr-operator/<package-slug>/replies/<reply-id>.md. Same path as v1 — downstream tooling does not branch on version.

Trigger /kai-sales-meeting-prep when the reply implies a meeting, demo, call, referral intro, or buying conversation.

Escalate when

• The reply is legal, regulated, sensitive, hostile, ambiguous, or high-value — category needs_human_review, no draft.
• Consent basis or suppression state for the contact cannot be established.
• The reply disputes a claim made in the original outbound message.
• The safe next action requires a live send, CRM mutation, or calendar action and no approver is named.
• Reply content contradicts the source evidence the account was built on.