LinkedIn Agent

Daily LinkedIn draft agent, built as a Trigger.dev task. Per project, once a day: drafts a LinkedIn post with an LLM, runs deterministic brand-voice guardrails and a post-quality guard, then posts a Slack card for human-in-the-loop Approve/Deny.

Extracted from bowtiefunnel/agents (formerly agent/subagents/linkedin/), where it originated and still runs in production.

Stack

• Trigger.dev — scheduled task runtime (daily-linkedin-drafts-scheduled)
• OpenRouter — LLM drafting
• Supabase — project list, review cards, audit log, draft memory
• Slack — HITL Approve/Deny review surface
• Langfuse (optional) — LLM call tracing/cost observability

Setup

npm install
cp .env.example .env   # fill in real keys
npm run typecheck
npm test

This repo is currently unlinked from any Trigger.dev project (extracted for ownership/isolation, not yet cut over from prod). To run it standalone:

1. Create a new project in the Trigger.dev dashboard.
2. Set TRIGGER_PROJECT_REF / TRIGGER_SECRET_KEY in .env.
3. npm run dev (local) or set -a; . ./.env; set +a; npm run deploy (prod).

Layout

agent/
  subagents/linkedin/     the agent: daily-linkedin-drafts.ts, lib (linkedin-memory, post-guard)
  connections/            shared clients: openrouter, langfuse, slack, supabase
  lib/                    shared helpers: blocks (Slack card builder), company-context,
                           draft-memory, guardrails, trace
  schemas.ts               shared Zod schemas (Project, AgentPayload)

Relative import depth mirrors the parent repo's agent/ folder exactly, so nothing was rewritten during extraction.