Kelly inquiry

Kelly Inquiry

Overview

Use this skill as Kelly's inbound-inquiry reception desk for foreign-trade sales. Inquiries arrive via WhatsApp, Instagram, Messenger, and email; sales reply late and leads leak. The skill aggregates them into one file-backed App-in-Skill pipeline (new → replied → quoted → negotiating → won/lost), maintains a product knowledge base + FAQ so the agent can draft accurate replies and quotes, enforces approval before anything is sent, and surfaces stale deals past the follow-up SLA.

Default interaction mode: App UI. Unless the user explicitly asks for chat-only handling, check onboarding/config, refresh or load the local inquiry snapshot, start/reuse the local app with app/start.sh, and give the actual local URL. Use chat-only mode only when the user says "纯聊天", "chat only", "不要打开 UI", or similar.

App UI Screenshots

<table> <tr> <td width="50%"><img src="assets/screenshots/overview.webp" alt="Kelly Inquiry overview"></td> <td width="50%"><img src="assets/screenshots/approvals.webp" alt="Kelly Inquiry approvals"></td> </tr> <tr> <td><strong>Overview</strong><br>Inquiry command desk with reply SLA counters, weekly channel mix, pipeline funnel, and stale-deal alerts.</td> <td><strong>Approvals</strong><br>Approval-gated outbox for replies and quotes — nothing is sent until reviewed.</td> </tr> <tr> <td width="50%"><img src="assets/screenshots/inquiries.webp" alt="Kelly Inquiry pipeline"></td> <td width="50%"><img src="assets/screenshots/quotes.webp" alt="Kelly Inquiry quotes"></td> </tr> <tr> <td><strong>Pipeline</strong><br>Inquiries across WhatsApp, Instagram, and email with country, stage, value estimate, and next follow-up.</td> <td><strong>Quotes</strong><br>Quote worksheets with line items sourced from the product KB, validity, and min-price guards.</td> </tr> </table>

Boundary

• The app reads and writes local files only and never touches any network beyond 127.0.0.1. It cannot send anything: the composer queues drafts, the approvals view records verdicts.
• Every outbound reply AND quote is approval-required. The skill executes only items whose approvals status is approved, via scripts/send_approved.ts (dry-run by default), and never bypasses the dry-run → --send sequence.
• Own accounts only: read and send exclusively through accounts the user owns and has configured. Respect each platform's terms of service and rate limits; prefer official APIs; keep collection read-only.
• Never store passwords, QR-login payloads, or session tokens. For browser_agent collection the agent drives the user's own already-authenticated web session and stores only inquiry text needed for review.
• Product and pricing data stays local. Treat customer contacts, conversation excerpts, price floors, and quotes as sensitive: do not commit config.local.json, env files, app/.data/, exports, tokens, or customer PII.

First Run And Onboarding

On invocation, check app/.data/onboarding.json and private config readiness. If onboarding is absent/incomplete, guide setup before collecting real inquiries.

Private config priority:

1. KELLY_INQUIRY_CONFIG=/absolute/path/to/config.json
2. skills/kelly-inquiry/config.local.json
3. ~/.config/kelly-inquiry/config.json
4. skills/kelly-inquiry/config.example.json as template only

Env priority:

1. Existing environment variables
2. KELLY_INQUIRY_ENV_FILE=/absolute/path/to/.env
3. Repository root .env
4. skills/kelly-inquiry/.env.local
5. ~/.config/kelly-inquiry/.env

Onboarding asks, turn by turn:

1. Which channels receive inquiries (WhatsApp / Instagram / Messenger / email) and which connector method per account (see Collection Workflow). Ask for non-secret details only: channel, display name, handle, and which env var names hold the tokens. Never ask the user to paste secret values into chat; secrets belong only in local env files.
2. Product KB import: a JSON or CSV file of products (SKU, MOQ, price range incl. price_min floors, lead time, specs, FAQ), imported via scripts/sync_products.ts.
3. Quote defaults: currency, validity days, incoterm/payment terms, and whether the min-price guard is enabled.
4. Follow-up SLA days per stage (defaults: new 1, replied 2, quoted 3, negotiating 5).
5. Reply style: tone, language policy, signature, and "do not say" guardrails.

When setup is complete and the user confirms, write app/.data/onboarding.json:

{
  "completed": true,
  "completed_at": "ISO timestamp",
  "config_version": "1"
}

Local App

Start the app with:

skills/kelly-inquiry/app/start.sh

The app uses local HTTP on 127.0.0.1, preferring ports 3000 through 4000, or KELLY_INQUIRY_UI_PORT when set. /api/state identifies the app as kelly-inquiry.

Required app views (hash routes):

• #/overview: inquiry command desk. Human-attention panel (replies/quotes awaiting approval, unanswered new inquiries, stale deals past the follow-up SLA), KPI cards (inquiries this week with channel badges, reply median time, quotes sent, win rate), a pipeline funnel summary (inline SVG bars: new→replied→quoted→negotiating→won), and an oldest-unanswered indicator.
• #/inquiries and #/inquiries/<id>: the pipeline table — customer, country flag/code, channel badge, product interest, stage, value estimate, last message age, next follow-up date, owner. Detail: conversation excerpt (bubbles), customer profile (company, country, source), linked products, quote history, an agent-drafted reply with editable text plus Queue reply (goes to Approvals), and a follow-up scheduling field.
• #/quotes and #/quotes/<id>: quote worksheet — quote no, customer, line items, currency, validity, status (draft/sent/accepted/expired/declined). Detail: editable line items sourced from the product KB (draft quotes only), terms, and agent pricing notes with the min-price guard result from the KB floors.
• #/approvals: the review queue with workflow states needs_review / changes_requested / approved / done / blocked over outgoing replies AND quotes. Each item shows target channel + customer, the draft, reason/context, editable text, decision buttons (approve / request changes / save edit / block), and stable refs (Reply #1 / Quote #2). done means sent, with the execution result from the execution report.
• #/products and #/products/<id>: the product KB — cards with name, SKU, MOQ, price range, lead time, FAQ count; detail with specs and the FAQ entries the agent uses for drafting.
• #/settings: sanitized config — channels/accounts with connector method + env readiness booleans, product KB source, quote defaults (currency, validity days, min-price guard), follow-up SLA, data provider, onboarding state, sync log, and last execution report. Never secrets.

Demo mode:

• ?demo=overview, ?demo=inquiries, ?demo=quotes, ?demo=approvals, and ?demo=detail (opens the featured hot WhatsApp inquiry wa-mueller-led-panels with a drafted reply and a draft quote) select named deterministic mock scenes.
• lang=en or lang=zh forces UI chrome language for screenshots. With lang=zh the chrome AND agent-generated content (reasons, notes, product names, pricing notes) are localized; drafted replies and conversation quotes stay in the buyer's language. Deep links such as /?demo=detail&lang=zh#/inquiries/wa-mueller-led-panels work.
• Demo mode never reads or writes app/.data/. Composer, approvals, follow-up, and quote-edit buttons still work but act on in-memory state only and show a demo notice.

UI language: English and Chinese chrome with Auto default following the browser language; explicit selector persisted locally. Keep customer names, message content, and drafted outbound text in their original language.

File Contract

Read references/inquiry-schema.md before editing the app, scripts, or any generated JSON.

• app/.data/inquiry_snapshot.json: accounts, inquiries (with conversation excerpts), quotes, products (KB), approvals (batch items), metrics, sync log. Written only by the scripts and the app server's queue/decision/follow-up/quote endpoints.
• app/.data/decisions.json: user decisions and notes keyed by approval item id.
• app/.data/agent_tasks.json: queued agent work — revise_reply / revise_quote tasks from request_changes decisions, plus follow_up tasks for stale deals.
• app/.data/execution_report.json: latest send run results.
• app/.data/onboarding.json: onboarding completion marker.
• app/.data/agent.lock: temporary lock while the skill ingests, syncs, or sends. While it exists the app rejects writes (423) and renders the composer, approvals, follow-up field, and quote editor read-only.

Validate with node scripts/validate_ui_schema.ts before relying on a snapshot in the UI.

Collection Workflow

1. Detect mode. Default to App UI.
2. Load private config. If only config.example.json exists, enter onboarding.
3. Collection reuses the connector reality documented in kelly-messenger. Per account, declare a connector:
   • whatsapp_cloud — WhatsApp Business Cloud API (access_token_env + phone_number_id_env). Inbound messages arrive via webhook only, so history is collected via ingest payloads; sends use the Cloud API.
   • instagram_graph / messenger_graph — Meta Graph API for professional-account DMs and Page messages (access_token_env + ig_user_id_env / page_id_env).
   • email_agent — hand off to the kelly-email skill: it collects inquiry emails, the agent normalizes them into an ingest payload; sends go back through kelly-email drafts.
   • browser_agent — the agent drives the user's own already-authenticated web session (e.g. WhatsApp Web, Instagram web) with the browser skill, then writes a payload. No passwords or QR secrets are ever stored.
   • manual — the user or agent prepares an ingest payload by hand.
4. All collected data enters through one write path: node scripts/ingest_inquiries.ts payload.json. It validates the payload, dedupes by stable inquiry/message ids, merges into the snapshot, applies the stage heuristic (an outgoing reply promotes new → replied), updates the account card and sync log, honors the agent lock, and recomputes metrics.
5. While drafting, the agent may attach a suggested_reply per inquiry (prefilled in the composer) and queue reply/quote drafts into snapshot.approvals[] with suggested_by: "agent" and a clear reason — always grounded in the product KB and reply style, never below price_min.
6. Start/reuse the UI and report the URL. Surface connector problems as snapshot warnings, not silent failures.

Quoting Workflow

1. Ground every quote in the product KB: SKU, MOQ, tier pricing inside price_min–price_max, lead time, and FAQ facts (certificates, OEM options, dimming, packaging). Import/refresh the KB with node scripts/sync_products.ts products.json|products.csv (zero-dependency CSV parser with quoted-field support).
2. Min-price guard: config quote_defaults.min_price_guard plus per-product price_min floors. Any line priced below its floor raises a pricing_alerts entry; with block_below_price_min the agent must not queue such a quote for sending — block it and ask the user instead.
3. Build the quote as a draft in snapshot.quotes[] (quote no, line items, currency, validity from validity_days, terms from quote defaults, pricing notes explaining the tier used and the guard result) and queue a matching kind: "quote" approval item referencing it.
4. The user edits draft line items in #/quotes/<id> (the server recomputes totals and re-runs the guard) and gives the verdict in #/approvals.
5. After a quote is sent, set its status to sent; track accepted / expired / declined from the conversation, and move the inquiry stage accordingly (quoted, negotiating, won, lost).

Approval And Send Workflow

scripts/send_approved.ts is the executor — there is no separate execute_decisions.ts.

1. Queue: the user writes or edits a reply in the composer (optionally starting from the agent's suggested_reply) and clicks Queue reply; the app appends it to snapshot.approvals[] as needs_review. The agent queues its own reply/quote drafts the same way.
2. Review: in #/approvals the user approves, edits (Save edit), requests changes, or blocks each item. Decisions are mirrored into decisions.json; request_changes enqueues a revise_reply/revise_quote task in agent_tasks.json.
3. Agent revision loop: poll agent_tasks.json, redraft the text honoring the comment, the config reply_style, and the KB (re-check the min-price guard for quotes), set the item back to needs_review, and mark the task done.
4. Send: only after the user asks to send, run node scripts/send_approved.ts (dry-run) and show the plan — planned sends, targets, and missing-token blockers. With explicit approval, run node scripts/send_approved.ts --send: it re-checks the lock and each item's approval immediately before sending, sends API-connector items via the official APIs (WhatsApp Cloud, Instagram/Messenger Graph), marks email_agent/browser_agent/manual items as handoff_to_agent for the agent to deliver (kelly-email drafts, or the user's own web session), sets sent items to done, and writes execution_report.json.
5. Report per-item results back to the user with the stable Reply #N / Quote #N refs.

Follow-Up Reminders

Stale deals are how leads leak; the agent owns catching them.

1. On every invocation (and after every ingest), compare each active inquiry (new/replied/quoted/negotiating) against follow_up.sla_days for its stage and its next_follow_up date.
2. For each overdue deal, append a follow_up task to agent_tasks.json (inquiry id, quote id if one expired, and a comment explaining which SLA was exceeded) so the queue survives interruption.
3. Draft the follow-up reply into the approvals batch (suggested_by: "agent", reason citing the SLA breach and deal value), then mark the task done — the human still approves before anything is sent.
4. The UI surfaces the same signal: the sidebar stale-deal counter, the overview "Stale deals past follow-up SLA" panel, and red overdue dates in the pipeline table. The user can reschedule with the follow-up field in the inquiry detail (/api/inquiries/followup).
5. When the user asks "what should I chase today", answer from agent_tasks.json + the stale list, with Reply #N refs for anything already drafted.

Safety Defaults

• Never send without an approved status recorded in the approvals batch, and never bypass the dry-run → --send sequence.
• Never quote below a product's price_min without an explicit human decision recorded on the item.
• Prefer read-scoped tokens where the platform offers them; keep collection strictly read-only.
• Redact tokens and token-like strings from logs, reports, and UI state; expose only env-var readiness booleans.
• Keep sends idempotent: stable item ids, execution results stored on the item, and re-reading approvals before each send.
• If a send target is missing (provider_conversation_id), block that item and ask for configuration instead of guessing.
• Honor platform rate limits; on 429s back off rather than retrying aggressively.