Skip to content

AI engineering for teamsOwn product

AI Agents Mark Work Done Without Proof: Coding-Agent Pipelines With Verdict Files and Sandboxes

I build coding-agent workflows with isolated execution, explicit verdict files and review before a task is accepted.

For
CTOs and development teams using coding agents
Checked

Sounds familiar?

Tick what applies to you

The situation AI engineering for teams
Your agent has finished talking. The work has not passed acceptance.

Tick the lines that describe your case.

Describe my task

How I solve it

What I build

I build a pipeline that treats acceptance as a separate step. An agent works within defined access and an isolated environment. Its output includes a verdict file tied to the code and checks. The supervisor records incomplete work as incomplete, and review happens before acceptance.

The process

How it works

5 steps. Scope and a fixed price are agreed before the first one.

  1. Define the task contract, allowed access and required evidence.

  2. Run the task in an isolated workspace with explicit resource limits.

  3. Capture changes, check results and a handover for interruptions.

  4. Reject acceptance when the verdict or required evidence is absent.

  5. Review the result and test the workflow on both passing and failing tasks.

Handover

What you get

  • A task contract and acceptance gate for your repository.
  • Traceable results and a path for failed or interrupted runs.
  • An operating runbook with access and review boundaries.

Built with

  • An agent runner
  • isolated workspaces
  • a sandbox where appropriate
  • a local/cloud worker
  • check logs
  • verdict files and a supervisor
  • Existing upstream frameworks remain credited

Proof

Done before, with dates and numbers

Own product Built and run by me, in production.

Before → after

In my internal fleet work, chains could close without a verdict file. I added an explicit acceptance requirement rather than relying on the agent saying it was done. The related sandbox remains a pilot; this page does not promise unattended correctness for arbitrary tasks.

  • My internal agent tooling

    a fleet runner, a sandbox pilot and a local/cloud worker inform this workflow. The upstream agent frameworks are third-party software; my work is the execution policy, supervision and acceptance process.

What you can look at

A proposed synthetic run bundle: task contract, code revision, passing or failing check output, verdict and reviewer decision. Preparing a public bundle is part of the review work still to agree.

Built for my own engineering work. The sandbox is a pilot. Broad autonomous certification of the local worker is not claimed. OpenCode and Oh My OpenAgent are upstream frameworks.

Price and timeline

What it costs and how it runs

A Custom Engineering Project starts from USD 1,500, with the first task class and acceptance checks agreed in writing. Continued engineering can use a Technical Partnership from USD 1,000 per month. Neither route promises unrestricted autonomous operation.

QuoteFixed before work starts

Price
Custom Engineering Project from USD 1,500. Ongoing work can use a Technical Partnership from USD 1,000 per month. Scope and price agreed in writing; discussing the task is free.
Timeline
Start with one repository and an agreed task class. Delivery dates and acceptance checks are fixed in the written scope.
First step
Describe the task. I reply within one working day, free, and tell you which option fits — or that you can fix it yourself.
Describe a task like this

A note from Daniilbefore you decide

When you don’t need this

If your team uses an agent interactively, reviews each change and already has reliable CI evidence, a written checklist may be enough. A pipeline is useful when repeated or unattended runs make missing evidence hard to spot.

— Daniil

Questions

What people ask about this

Is a verdict file enough by itself?

No. It must point to the relevant code version, executed checks and review evidence. Missing or failed evidence keeps the task open.

Can every task run without supervision?

No. I agree which tasks can run, what access they need and where a person must decide.

More like this

All situations
Daniil Maximkin

Hi, I’m Daniil.

I work with you from defining the problem to implementation and handover. You talk to the person who does the work. I work in English and Russian.

My story My work

Have a task like this?

Describe your task

The first answer is free, within one working day. Or write directly: next@taskfordaniel.com