Last updated
Vibe-Coded App Rescue
Fix My Codex-Built App
- From
- £1,000
- Deliverables
- 6
- Process
- 5 steps
- Studio
- Telford, UK
Built your app with OpenAI Codex and it stalls before production? We review the code and the pull requests Codex produced, replace mock-only tests with real ones, fix what is broken and set Codex up so your next change is safe.
In short
Circlemount, a Telford, Shropshire software team, fixes apps built with OpenAI Codex for UK founders and businesses. It reviews the code and pull requests Codex produced, replaces mock-only tests with real ones, fixes bugs, security gaps and deployment, and sets up AGENTS.md and approval settings. The initial app assessment is free; projects start from £1,000 + VAT.
5.0 Google (17 reviews)£57m client revenue213 clients

Codex can ship a pull request in minutes. Reviewing it is still your job.
OpenAI Codex is a coding agent that works in your repository from the command line, your editor, the desktop app or the cloud, where tasks run in their own environment and can open pull requests. It reads project instructions from AGENTS.md and works within the approval and sandbox settings you choose. We use it ourselves, alongside Antigravity, Claude Code and Lovable, to build our own products.
Problems appear when output outpaces review. Pull requests pile up and get merged because they look plausible. Tests pass because they only test mocks. A task is marked finished halfway through. Full access is switched on to save clicks, and the agent can then run commands well outside the project: in August 2026 OpenAI fixed a Codex bug in which a clean-up command deleted real user files.
We review the code and the pull requests Codex produced, replace tests that prove nothing with tests that check real behaviour, fix what is broken, and set Codex up so the next change is safe: clear AGENTS.md conventions, sensible approval and sandbox defaults, and review before merge.
Sound familiar?
01
“Codex marked it finished halfway”
Tasks closed as complete while parts of the feature are missing or never wired up.
02
“The tests only test mocks”
A green test suite that says nothing about whether the app actually works.
03
“Dozens of pull requests nobody reviewed”
Merged because they looked right, now tangled together and hard to undo.
04
“Full access was on”
An agent that could run anything, anywhere, and a worry about what it touched.
05
“Credits gone, app still broken”
Round after round of agent tasks that move the problem rather than fix it.
What We Deliver
01
Code and pull request review
A plain-English review of the merged code and open pull requests: what to keep, rework or close.
02
Tests that test something
Mock-only tests replaced with tests of real behaviour, so a green check means the feature works.
03
Root-cause fixes
Bugs, broken flows and half-finished features fixed properly, in reviewed branches.
04
Deployment and environments
Repeatable builds, secrets kept out of the repository, staging and production that match.
05
A safe Codex setup
AGENTS.md conventions, approval and sandbox settings that fit your risk, and a review routine for agent pull requests.
06
Security, speed and SEO
Database access rules, authentication, performance and search visibility checked and fixed.
How It Works
1. Share the repository
Read access to the code, the open pull requests and a description of what is wrong.
2. Free app assessment
We review the code, pull requests, settings and deployment, and reproduce the problems.
3. Proposal
What we will fix and in what order, the timeline and a fixed price.
4. Fix, test, deploy
Reviewed branches with real tests, a staging copy for you to check, then production.
5. Handover or support
A walkthrough of the new setup, or an ongoing support plan.
From pull requests to production
- AssessRead the code, open pull requests, AGENTS.md and approval settings.
- UntangleKeep what is understood, rework or close the rest.
- ProveReplace mock-only tests with tests of real behaviour.
- ShipStaging first, then a repeatable production deploy with rollback.
- GuardAGENTS.md conventions, safe defaults and review before merge.
Sources: OpenAI Codex documentation; The Decoder on the Codex file-deletion fix.
Building with agents, or rescuing what they built?
This page is for apps Codex has already built. To build agents of your own, or to compare coding agents, start here.
What it costs
The initial app assessment is free, project work starts from £1,000 + VAT, and ongoing support plans start from £125 a month + VAT.
01
Free app assessment
We review the code, pull requests and Codex settings, then send a written, tailored proposal. No obligation.
02
Projects from £1,000 + VAT
Fixes, features, security and launch work, priced for your app in a written proposal before anything starts.
03
Support from £125 a month + VAT
Monitoring, updates, fixes and help as you keep building, on a monthly plan you can stop.
Apps we've built with AI, running in production
We build with AI coding agents every day, so we know the habits they bring to a codebase.
The Study Room's Portal, Dentorb and ORBQ were built with AI-assisted development, using Antigravity, Claude Code, Codex and Lovable, and every change was reviewed, tested and owned by an experienced developer. We know where these tools are strong, where they cut corners, and what production asks of the code they write.
We Serve Businesses Across the UK
Based in Telford, we provide Fix My Codex-Built App to businesses nationally including:
Frequently Asked Questions
Can you fix an app built with OpenAI Codex?
Yes. Circlemount fixes apps written with Codex for UK founders and businesses. We review the merged code and open pull requests, replace mock-only tests with real ones, fix bugs, security gaps and deployment, and set up AGENTS.md conventions and approval settings so your team can keep using Codex safely.
Why did Codex stop before the work was finished?
Agents judge progress from what they can see. On long tasks they can lose the original goal, treat passing tests as proof even when the tests only check mocks, and end the turn early. Smaller tasks, clear acceptance criteria in AGENTS.md and tests of real behaviour make finishing verifiable.
Which Codex approval and sandbox settings are safe?
For most teams, start with approval for anything outside the project and a workspace-write sandbox, and keep full access for isolated environments. Production credentials should never be available to the agent. We set defaults that match your risk and document them so everyone works the same way.
Codex deleted files. Can you help?
Yes. First we stop further changes, recover what version control, backups or your host can restore, and check what else the session touched. Then we fix the cause: narrower sandbox and approval settings, credentials kept out of reach, and a review step before agent changes reach shared or live environments.
Can we keep using Codex after you fix the app?
Yes. We leave AGENTS.md instructions that describe the project and its conventions, tests that make agent pull requests easy to judge, and a review routine. Your team keeps Codex's speed for everyday work while the codebase stays tested, understood and safe to deploy.
Do you use Codex yourselves?
Yes. Our developers use Codex alongside Antigravity, Claude Code and Lovable to build products such as The Study Room's Portal, Dentorb and ORBQ, and every change is reviewed, tested and owned by an experienced developer. Circlemount is independent and not an OpenAI partner.
How much does it cost to fix a Codex-built app?
The initial app assessment is free: we review the code, pull requests and settings and send a tailored proposal. Project work starts from £1,000 + VAT, depending on what is broken and how large the app is, and ongoing support plans start from £125 a month + VAT.





