SalesTouch

Build me a local LinkedIn outreach cockpit inspired by SalesTouch. Requirements:

- Node 22 + Express + better-sqlite3; one localhost web app with no accounts.
- Import prospects from CSV with name, company, role, LinkedIn URL, source,
  notes, and last-contact date; validate and deduplicate by LinkedIn URL.
- Let me paste profile, company, post, and conversation text into each record.
  Never crawl LinkedIn or read browser cookies.
- Store my offer and ICP rules in config.json. Use the OpenAI API to return a
  fit score, evidence, a personalized angle, a connection note under 300
  characters, a first DM, and one follow-up. Put OPENAI_API_KEY in .env.
- A kanban pipeline: new, researched, ready, contacted, replied, won, lost.
- A Today queue showing due follow-ups and a configurable manual daily limit.
- Each action opens the LinkedIn profile in a new tab and has copy buttons for
  the approved text. It must never click, send, invite, comment, or publish.
- Let me paste a new reply, show the complete conversation history, and draft
  a suggested response for approval.
- Log every status change and copied draft in SQLite; export prospects and
  activity as CSV. Never pretend a copied message was sent.
- Keep all data local. No cloud database, telemetry, background workers, or
  multi-user features.
- Explicitly out of scope: LinkedIn login or cookies, scraping, Sales Navigator
  automation, residential proxies, auto-sending, engagement, publishing,
  multi-account orchestration, and claims that this is account-safe automation.
- README: setup, CSV format, backup path, and an honest explanation that the
  tool replaces planning and drafting only, not SalesTouch's LinkedIn execution.

# AGENTS.md — Build guide for SalesTouch

## Project scope
Import or paste prospects, score them from user-supplied context, draft personalized outreach with an LLM, and track manual follow-ups in a local pipeline.

Catalogue verdict: no. The local CRM and AI drafting layer are straightforward, but SalesTouch's core value is reliable access to LinkedIn's private network: authenticated sessions, live data extraction, residential IP routing, human-paced queues, limits, cooldowns, reconnection, and ongoing adaptation to platform changes. A one-shot clone either stops at manual copy and paste or becomes brittle automation that can put the LinkedIn account at risk.
Use the implementation prompt below to define the deliverable. Complete each phase's acceptance checks before extending the scope.

## Working agreement
- Inspect the repository and its existing instructions before choosing paths, dependencies or commands. Keep one coherent stack and explain changes to the proposed architecture.
- Plan a vertical slice that accepts a real input and produces the useful output described below. Persist only the state the prompt calls for; respect memory-only and upstream-managed workflows. Use fixtures only when they are clearly labelled.
- After scaffolding, document the actual install, development, check and build commands in README and keep them synchronized with the package or project manifest. Do not report commands as successful unless they ran.
- Work in small steps. At handoff, list implemented flows, checks actually performed, remaining blockers, and any credentials or provider setup the owner must supply.
- Do not publish, spend money, contact customers, delete source data or run irreversible migrations without the project owner's authorization.

## Prerequisites
- OpenAI API key
- LinkedIn account
- manual profile and conversation input

## Stack and architecture
- Node 22 + Express + better-sqlite3; one localhost web app with no accounts.
- Model workflows, triggers, step executions, retries, and external receipts; keep stable source IDs and timestamps.
- Interface for this LinkedIn outreach automation workflow: a local web page with input, progress, review, and export views.

## Security and data integrity
- Import prospects from CSV with name, company, role, LinkedIn URL, source, notes, and last-contact date; validate and deduplicate by LinkedIn URL.
- Keep API keys and OAuth secrets in .env; redact them from logs, generated files, and error responses.
- State the practical limit: live LinkedIn and Sales Navigator searches and audience extraction.

## Agent implementation rules
- Project rule, data: Model workflows, triggers, step executions, retries, and external receipts; keep stable source IDs and timestamps.
- Project rule, behavior: Import prospects from CSV with name, company, role, LinkedIn URL, source, notes, and last-contact date; validate and deduplicate by LinkedIn URL.
- Project rule, recovery: Import prospects from CSV with name, company, role, LinkedIn URL, source, notes, and last-contact date; validate and deduplicate by LinkedIn URL.

## Optional agent skills and references
No installable skill is listed for this guide; use the project rules and the stack documentation.

Read the linked SKILL.md and its dependencies before adding a skill. Select only the skills matching this project's runtime and task; their documentation does not supply API access, credentials or approval to perform external actions. Pin the reviewed revision where the tool supports it. Follow the chosen agent's documented project-level installation mechanism.

## Distribution ideas
These are optional planning notes. Obtain the owner's approval before publishing or contacting anyone.
- Demonstrate this LinkedIn outreach automation operation with sample or user-owned data: Import prospects from CSV with name, company, role, LinkedIn URL, source, notes, and last-contact date; validate and deduplicate by LinkedIn URL.
- Show a concrete result or export: Log every status change and copied draft in SQLite; export prospects and activity as CSV. Never pretend a copied message was sent.
- State the practical limit: live LinkedIn and Sales Navigator searches and audience extraction.

## Engineering roadmap
1. Phase 1, architecture and data: Node 22 + Express + better-sqlite3; one localhost web app with no accounts. Model workflows, triggers, step executions, retries, and external receipts; keep stable source IDs and timestamps.
2. Phase 2, implement: Import prospects from CSV with name, company, role, LinkedIn URL, source, notes, and last-contact date; validate and deduplicate by LinkedIn URL.
3. Phase 3, implement: Store my offer and ICP rules in config.json. Use the OpenAI API to return a fit score, evidence, a personalized angle, a connection note under 300 characters, a first DM, and one follow-up. Put OPENAI_API_KEY in .env.
4. Phase 4, review and output: A kanban pipeline: new, researched, ready, contacted, replied, won, lost. Log every status change and copied draft in SQLite; export prospects and activity as CSV. Never pretend a copied message was sent.
5. Phase 5, recovery and acceptance: Import prospects from CSV with name, company, role, LinkedIn URL, source, notes, and last-contact date; validate and deduplicate by LinkedIn URL. Verify this invariant with a saved fixture: A retry after a provider timeout must not repeat a completed external action without an explicit reconciliation step. State the practical limit: live LinkedIn and Sales Navigator searches and audience extraction.

## Paid-product capabilities outside this build
- live LinkedIn and Sales Navigator searches and audience extraction
- authenticated messages, invitations, engagement, and publishing
- residential IP routing and resilient LinkedIn sessions
- human-paced queues, limits, cooldowns, and account safety controls
- multi-account orchestration, activity logs, and analytics

## Implementation prompt
Build me a local LinkedIn outreach cockpit inspired by SalesTouch. Requirements:

- Node 22 + Express + better-sqlite3; one localhost web app with no accounts.
- Import prospects from CSV with name, company, role, LinkedIn URL, source,
  notes, and last-contact date; validate and deduplicate by LinkedIn URL.
- Let me paste profile, company, post, and conversation text into each record.
  Never crawl LinkedIn or read browser cookies.
- Store my offer and ICP rules in config.json. Use the OpenAI API to return a
  fit score, evidence, a personalized angle, a connection note under 300
  characters, a first DM, and one follow-up. Put OPENAI_API_KEY in .env.
- A kanban pipeline: new, researched, ready, contacted, replied, won, lost.
- A Today queue showing due follow-ups and a configurable manual daily limit.
- Each action opens the LinkedIn profile in a new tab and has copy buttons for
  the approved text. It must never click, send, invite, comment, or publish.
- Let me paste a new reply, show the complete conversation history, and draft
  a suggested response for approval.
- Log every status change and copied draft in SQLite; export prospects and
  activity as CSV. Never pretend a copied message was sent.
- Keep all data local. No cloud database, telemetry, background workers, or
  multi-user features.
- Explicitly out of scope: LinkedIn login or cookies, scraping, Sales Navigator
  automation, residential proxies, auto-sending, engagement, publishing,
  multi-account orchestration, and claims that this is account-safe automation.
- README: setup, CSV format, backup path, and an honest explanation that the
  tool replaces planning and drafting only, not SalesTouch's LinkedIn execution.

## Completion evidence
Demonstrate the prompt's acceptance scenarios against the scoped workflow. Include setup from a clean checkout and failure recovery. Check persistence across restart and export/restore only for the state the prompt says to store; for memory-only tools, confirm that temporary content is discarded as specified. Record actual results and remaining limitations. A detailed plan alone does not establish a working replacement.