Claude Automation for B2B Growth

Write project instructions that give Claude Code the product architecture, coding conventions, and validation commands.

Use Claude Code to explore an unfamiliar part of the repository before proposing a change.

Use Claude Code to plan a bounded feature from its acceptance criteria.

Implement a bounded feature with Claude Code on an isolated Git branch using small, reviewable commits.

Validate an AI-assisted change with formatting, linting, type checking, and automated tests.

Review an AI-generated diff for correctness, security, and unintended scope.

Diagnose a failing feature by giving Claude Code targeted logs, tests, and reproduction steps.

Containerize the FastAPI application with a reproducible production build.

Create versioned Supabase migrations for organizations, users, memberships, and roles.

Connect a Next.js login and session flow to Supabase Auth.

Validate a Supabase access token and derive the authenticated user context in FastAPI.

Resolve and authorize the active organization for each authenticated API request.

Write and test row-level security policies that prevent cross-tenant data access.

Enforce and test role-based permissions for owners, operators, reviewers, and viewers.

Implement a client workspace onboarding flow.

Record privileged actions in a tenant-scoped audit log.

Send a typed request from FastAPI to the Claude API.

Write a prompt contract that separates instructions, tenant context, source evidence, and user input.

Validate Claude responses against a Pydantic schema.

Implement a Claude tool call with an explicit input schema and allowlist.

Reject tool calls that exceed the authenticated user's tenant or role permissions.

Apply input delimiting and data minimization to reduce prompt-injection and privacy risks.

Add timeout, retry, and idempotency-safe error handling to a Claude-powered endpoint.

Meter model, token, latency, and estimated cost data by tenant and enforce a configurable Claude usage budget.

Store and rotate a tenant's integration credentials in encrypted server-side storage without exposing them to the frontend.

Create and update GoHighLevel contacts through the connector layer.

Create and update GoHighLevel opportunities through the connector layer.

Verify, persist, and acknowledge an incoming webhook idempotently.

Map and batch-import Airtable lead records into the canonical Supabase data model.

Pass connector contract tests for pagination, rate limits, and transient API failures.

Persist automation-run, step, and connector-attempt statuses with actionable error details.

Run long-running, idempotent automation jobs in an Amazon SQS-backed worker with dead-letter handling.

Represent each tenant's ideal customer profile as versioned configuration.

Normalize company domains and contact identifiers before enrichment.

Deduplicate an incoming prospect against existing tenant records.

Retrieve approved public web sources for a prospect while respecting source-access rules.

Extract company attributes from retrieved evidence into a validated schema.

Attach source URLs and evidence snippets to every AI-inferred prospect attribute.

Calculate freshness and completeness indicators for an enriched prospect profile.

Route low-confidence or conflicting enrichment results to human review.

Translate an ideal customer profile into an explicit qualification rubric.

Calculate deterministic qualification features from enriched prospect data.

Use Claude to assess qualitative criteria against cited prospect evidence.

Combine deterministic and qualitative results into an explainable lead score.

Make the qualification workflow abstain when evidence is insufficient.

Map score bands to GoHighLevel pipeline stages and owner-routing rules.

Capture a reviewer correction as structured qualification feedback.

Measure qualification precision and recall on a labeled test set.

Apply tenant-configured consent, suppression, and contact-frequency rules before outreach.

Create a reusable brand-voice and value-proposition profile for a tenant.

Generate an outreach brief from prospect evidence and campaign intent.

Draft a personalized message whose claims are traceable to approved evidence.

Generate a multi-step follow-up sequence with explicit timing and stop conditions.

Run automated checks for unsupported claims, prohibited language, and missing personalization.

Implement a human approval step before activating an outreach sequence.

Classify an inbound reply and select the permitted next action or human-escalation path.

Convert a business objective into a structured campaign brief.

Create channel-specific content requirements for email, social, and landing-page assets.

Generate a content asset from an approved brief and tenant brand profile.

Transform one approved source asset into a channel-specific derivative.

Store content drafts, review comments, approvals, and versions.

Generate GoHighLevel campaign configuration from an approved sequence.

Add consistent UTM parameters and campaign identifiers to outbound links.

Schedule or publish only approved assets through a supported campaign connector.

Build a tenant-aware dashboard shell in Next.js.

Create a typed, authenticated Next.js data-access layer for the FastAPI API.

Display a filterable lead and opportunity pipeline for the active tenant.

Build an approval queue for enrichment, qualification, outreach, and content tasks.

Display automation-run status, retry state, and error details.

Implement a canonical event schema for acquisition, engagement, conversion, and revenue events.

Compute funnel conversion, campaign cost, revenue, and attribution metrics using a documented attribution rule.

Visualize funnel performance and campaign attribution in the client dashboard.

Create a versioned metric snapshot for use in growth analysis.

Build a retrieval step that supplies Claude only the relevant metrics and campaign context.

Generate a growth recommendation that cites the metrics supporting it.

Reject a recommendation containing a claim unsupported by the supplied metrics.

Rank candidate recommendations with an explicit impact, confidence, and effort rubric.

Convert an approved recommendation into an experiment brief with a hypothesis, success metric, guardrails, and tracking identifier.

Create a regression set for research, qualification, outreach, and recommendation outputs.

Use reviewer feedback and regression results to decide whether a prompt change is releasable.

Build, tag, and push reproducible API and worker images to Amazon Elastic Container Registry.

Store and rotate production runtime credentials in AWS Secrets Manager using least-privilege access.

Provision an ECS Fargate API service behind a TLS-enabled Application Load Balancer using infrastructure as code.

Provision an ECS Fargate worker service with least-privilege SQS access, autoscaling, and dead-letter handling.

Deploy the Next.js client application with AWS Amplify Hosting.

Create a CI/CD workflow that tests and deploys an approved revision.

Configure centralized logs, service metrics, and alerts for API failures and failed automation runs.

Execute and document launch drills covering tenant isolation, backup restoration, data export or deletion, and rollback.