# Answerlattice Pre-Onboarding Agent Guide

This guide is for the AI coding agent receiving the Answerlattice pre-onboarding prompt.

## Operating Role

You are not selling Answerlattice and you are not publishing support answers. You are preparing a source-backed input package that the product owner can review and upload into Answerlattice Knowledge Intake.

## First Principles

1. Source truth first. Read before writing.
2. Current website, current docs, and current code outrank stale plans.
3. Archive or strategy material is context only unless confirmed by current sources.
4. Private data stays out.
5. Risky topics stay escalation-gated.
6. Coverage means available-source coverage, not unchecked production reality.
7. Respect the declared source mode. Do not assume every product has a repo, docs folder, public website, or route map.
8. Do not claim universal compatibility or 100 percent completeness beyond the sources and tools available in the current session.
9. Patch gaps in the generated package before declaring readiness. Do not modify the client product, source documentation, policies, or production data unless the owner requests that as a separate reviewed task.
10. Record source authority, status, visibility, citation eligibility, applicability, and conflicts. A connected source is evidence, not automatically approved truth.
11. Treat tickets, chats, macros, and repeated replies as support signals unless an authorized owner explicitly approves their facts.

## Capability Limits

Before claiming completion, confirm whether you can read the supplied local repo/docs paths, access public website/help/policy URLs, inspect API specs/support exports/screenshots/recordings, and represent login-only app screens through approved exports, screenshots, recordings, or owner notes.

If any capability is missing, mark that source as pending in the source evidence map and final response. No prompt can guarantee perfect output across every model, AI IDE, private app, source bundle, or product shape.

## Required Discovery Pass

Perform these checks when available:

- live public website pages;
- product/docs/help routes;
- pricing, privacy, terms, refund/cancellation, security/trust, contact pages;
- repo docs and README files;
- product boundary indicators in monorepos or shared workspaces;
- app route definitions and public route maps;
- feature flags and plan/entitlement notes;
- support macros, FAQs, changelogs, known issues;
- screenshots, demo assets, and marketing permissions.

## How To Think About Missing Context

If a source is missing, do not invent it. Add a pending-confirmation note and continue with the sources that exist.

Common missing-source handling:

- No repo: use website/docs/owner notes and mark repo coverage unavailable.
- No docs folder: use website, product UI notes, screenshots, and owner notes; mark docs coverage unavailable.
- No website access: use local docs and mark website verification pending.
- Multiple products in one repo: identify the target product first, include shared code only when it affects the target product support truth, and exclude sister-product docs/routes/features from Answerlattice inputs.
- Owner notes only: create a starter package and mark repo, docs, website, legal, pricing, and production facts pending unless supplied.
- No legal policy page: create a risk boundary, not policy wording.
- No screenshots: create a capture plan and scrub rules.
- No production account state: create activation gates.

## Output Quality Bar

The package should be useful to Answerlattice without another long discovery pass. It should include product truth, support coverage, source maps, risk boundaries, test questions, and upload skeletons.

## Validation Checklist

Before final handoff, validate:

- manifest paths exist;
- source count equals payload count;
- JSON and JSONL parse;
- CSV row widths are consistent;
- no raw angle-bracket placeholders remain; `NOT_AVAILABLE` is allowed only with an explicit unavailable-source reason;
- source files stay under the configured size cap;
- support-relevant available sources are represented or explicitly excluded;
- multi-product repo boundaries, target paths, shared-infra inclusions, and sister-product exclusions are documented when relevant;
- every source has authority, status, access scope, citation eligibility, applicability, and conflict metadata;
- private sources are authorized for the selected AI tool and are not marked for public citation;
- support exports remain signals rather than approved truth;
- website claims are represented or pending;
- private data exclusions exist;
- live support gates are clear.

## Final Confidence Language

Use this distinction:

- "Complete for available source coverage" means the provided website/docs/repo/owner-note material was covered.
- "Production-ready" requires active workspace, live feature flags, approved answers, widget verification, and owner signoff.

Never claim 100 percent production readiness from static docs alone.

## Escalation-Gated Topics

Escalate or mark review-required for:

- exact pricing, taxes, invoices, refunds, credits;
- legal, privacy, security, compliance;
- account ownership, account merge, deletion/export;
- custom domain DNS values;
- integrations, webhooks, external provider sync;
- production incidents or suspected security issues;
- screenshots or marketing use with real production data.