Codex Automation · £3,500 · 7 days
OpenAI Codex, properly installed and tuned for your repo.
You have an OpenAI account. You have GitHub repos. Codex can review every PR, triage every issue, and gate every CI run on quality checks — but only if it's configured for the way your team actually ships. I install it, write the AGENTS.md guidelines, wire the GitHub Action, and tune it for a week against your real PRs. You keep using it after I leave.
What you get
- Codex installed on every repo you nominate — up to 10 repositories included
- AGENTS.md authored to your team's actual code conventions, security policies, review priorities
- Automatic PR reviews enabled with custom severity thresholds (P1 / P2 / nice-to-have)
- Issue triage automation — new issues get auto-labelled, prioritised, and routed to the right reviewer
- CI gating: Codex review becomes a required check before merge, configurable per branch
- Alert-monitoring runbook: Codex watches your error stream and opens triage PRs automatically when a stack trace recurs
- Weekly tuning sessions during the 7 days, so the bot learns your team's actual style
- Handover documentation: how to maintain it, how to add repos, how to update guidelines — written for whoever owns this on day 8
Why this is worth doing properly
Codex out-of-the-box is fine. Codex configured against your codebase, your conventions, your security rules, your alert stream — that's a meaningful productivity multiplier on every PR for the next two years.
Most teams who turn it on get noisy reviews, file dozens of false positives, and switch it off within a fortnight. The 7-day sprint exists to cross that gap properly the first time, with someone who's done it before, so your team trusts the output from day one.
What gets configured
- PR review automation with your own severity rubric
- AGENTS.md guideline files (top-level + per-package nested)
- GitHub Action workflows for headless Codex CI runs
- Issue triage rules — auto-label, auto-assign, auto-prioritise
- Branch protection: Codex review required before merge on protected branches
- Cost guardrails — per-repo monthly budget caps so a runaway loop can't blow up your OpenAI bill
- Sentry / Honeycomb / Datadog hooks: when an error spikes, Codex opens a fix PR
- Per-team review preferences (a frontend team and a platform team should not get reviewed the same way)
What you need before we start
- An OpenAI account with API access (Tier 1 or above — Codex needs GPT-5.5 access)
- Admin access to the GitHub organisation — or a teammate who can grant the GitHub App during install
- A short list of repos to onboard (1–10), prioritised
- 30 minutes of one of your senior engineers' time on day 1 to walk me through your code conventions
Day-by-day
- Day 1 — discovery: code-conventions interview, security policies, repo prioritisation, AGENTS.md draft 1
- Day 2 — install: GitHub App + Codex Cloud configured on first 3 repos, PR review enabled, AGENTS.md committed
- Days 3–5 — tune: Codex reviews live PRs, I tune the AGENTS.md against false positives + missed issues. Daily sync with your tech lead.
- Day 6 — CI gating + remaining repos: branch protection wired, remaining repos onboarded, alert-monitoring hook live
- Day 7 — handover: walkthrough, runbook documentation, owner identified, deferred items logged for your roadmap
Terms
- £3,500 fixed. 50% upfront, 50% on delivery.
- If Codex isn't reviewing PRs cleanly on at least 5 nominated repos by day 7, you pay nothing.
- You own everything. AGENTS.md files live in your repos. The OpenAI account stays yours. No vendor lock-in.
- £600/mo optional maintenance — I monitor the configuration, review monthly false-positive trends, ship updates as Codex evolves.
Got an OpenAI account that isn't earning its keep yet?
Tell me which repos you'd want reviewed first and what your code-review pain looks like today. I'll tell you within 24 hours whether the 7-day sprint is the right shape.