@nifinet - Codex Signal-Based Outbound Engine

Open original

@nifinet on X A Sortlist exec walks through building a daily outbound pipeline on OpenAI's Codex CLI, using ChatGPT sign-in instead of a separate API key. Six file-based Codex jobs sense buying signals, score accounts, draft and check messages, log outcomes, and learn weekly from replies.

  • Codex runs each step through codex exec against plain files (CSV in, JSON schemas out), with cron driving a morning shortlist and approval queue.
  • The author argues timing beats volume: outbound should trigger on real events like a role opening, a funding round, or a team change, not on cold lists.
  • Writing and checking are split into separate steps, and scheduled runs avoid approval prompts by keeping to files, shell, and APIs rather than interactive MCP calls.
  • A weekly learning step reads outcomes and recommends retiring weak triggers and banning phrases, so the engine gets more selective over time.

The author, who works on outbound at Sortlist, built a daily sales-signal pipeline that runs on Codex. Instead of paying for an OpenAI API key, the reasoning steps run through a ChatGPT/Codex plan using `codex exec`, with a cron job kicking off the chain every morning. The core idea is that outbound should start from a trigger, such as a role opening, a funding round, a leadership change, or a visible business problem, rather than cold prospecting. The system follows six stages around one shared account record: sense (clean raw inputs into concrete signals), score (rank accounts by whether a human should spend time on them today), write (draft messages only for high-scoring accounts, with the signal shaping the copy), check (a separate pass for weak claims or fluff), log (record outcomes along with the reasons behind them), and learn (a weekly review that tells the team which triggers to trust less). The build is deliberately plain: a folder with an AGENTS.md file for house rules, JSON schemas so each step's output stays predictable, prompt files, and a morning script that writes a shortlist of who moved, why now, what to say, and what happened last time. The author is firm about a few things. Start by running the process by hand on about 20 accounts before automating anything. Keep a human approval queue rather than auto-sending. Set the draft threshold high (70) so the goal is a handful of messages worth sending. Credentials for Codex live in a file that should be treated like a password, and external data sources such as a CRM, job boards, or funding feeds still have to be connected separately. The method stresses that the sense step should discard weak signals, and that the most useful habit is reading the single "why now" sentence to check whether a score deserves trust.

Ask about this article