Show and tell: a check step after the agent node — so a flow's code output gets reviewed before it ships

[Fikpeze123]

Hi all — I'm Frank, founder of Pactlyst.ai. I build in the open with AI help, and this is a small pattern I've been wiring into agent flows that I wanted to show here, because Langflow is where it reads most naturally.

The problem: a flow finishes, the output looks fluent, and it moves on. Nothing ever checked it.

So I add one step between the agent node and whatever publishes, saves, or sends the result — an API Request step that sends the finished work to Pactlyst Work Check:

POST https://api.pactlyst.ai/functions/v1/work-check
Authorization: Bearer $PACTLYST_KEY

{
  "work_text": "def get_user(id):\n    return db.execute(\"SELECT * FROM users WHERE id = \" + id)",
  "work_type": "code",
  "ask": "Write a helper that fetches one user by id.",
  "requesting_agent": "langflow-coding-flow"
}

That helper is the kind of thing a coding flow produces on a good day — and it's a SQL injection waiting for an id like 1 OR 1=1. The check reads the work against code playbooks (the same door also checks documents and software work) and returns a score, a verdict — Low risk, Needs review, or High risk — and a gate: Low risk flows on, Needs review waits for a named human to acknowledge the findings, High risk waits for a named human approver who isn't the requester. Every result carries a signed PactCheck Receipt™ and a public receipt link, so "the flow's output was reviewed" is something you can show later, not just remember. The work itself is never stored — hash and finding counts only.

To be plain about what this is: there is no official Langflow integration. It's one HTTP step you add yourself, with a free key from pactlyst.ai — ten minutes of wiring, and if your flow speaks MCP, the same check is a check_work tool call instead.

I'm opening a small design-partner circle for teams whose flows produce work somebody has to answer for — three months of Pro free, no card. Details in the docs at pactlyst.ai/docs. Mostly, I'd value this community's read: where in a flow would you put the check?

— Frank