How to use Codex to fix saved-airport and booking-link bugs
Codex repaired MatchDay’s conflicting airport memory and empty flight handoffs, using deletion, isolation and evidence gates before a human-authorized release.
devpost.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Reproduce this travel app’s saved-airport versus chat-memory conflict. Trace correction, deletion, reload and owner isolation; then verify provider handoffs against route, dates, party and freshness. Keep fixture, cached, live, available and bookable evidence distinct, and require human approval before release or any transaction.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Mohamed Zidan and osama zaidan
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The creators report a working release with memory and provider-handoff repairs and executed regression gates; no trip purchase, completed travel or independent inventory audit is claimed.
- Human involvement
- The owners defined correct memory, seller-link safety, risk approvals and release authority; Qwen runtime behavior is distinct from Codex engineering work.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- The owners supplied the YUL/YVR memory conflict and broken provider handoffs as observable failures.
- Codex repaired memory contracts and checked provider evidence, freshness and route-specific handoffs.
- Human-supervised execution used review and release gates, including deletion, reload and tenant-isolation checks.
Mohamed Zidan and osama zaidan used Codex CLI to build and verify MatchDay, a World Cup travel-planning product. A concrete failure gave the work direction: a screen displayed Montreal’s YUL as the saved home airport while chat claimed no durable preference and continued planning from Vancouver’s YVR.
Codex traced the conflicting memory paths and worked through correction, deletion, cross-session recall and tenant isolation. Another failure opened generic or empty flight-provider pages. The team turned that into a handoff contract covering route, dates, party, freshness and honest fallback behavior. Provider evidence was separated into fixture, cached, discovery, live, available and bookable states.
Humans supervised from phones, set product truth and risk boundaries, and authorized release. Codex implemented small slices, attempted to disprove them, collected test evidence and resumed an interrupted session from a durable handoff. The source reports backend, browser and database concurrency checks, including memory that stayed deleted after reload.
Codex was the engineering agent; Qwen Cloud runs the finished travel assistant. The outcome is a verified-by-the-creators product release and specific memory and link repairs, not a completed customer trip. Commercial transactions and an agency-facing product remain future directions. Reported test counts are not an independent assessment of travel inventory accuracy.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Devpost · Mohamed Zidan and osama zaidan
Frequently asked questions
Which agent runs the finished travel product?
Qwen Cloud runs the product; Codex CLI built and verified it.
What specific memory failure was repaired?
The interface saved YUL while chat planned from YVR and denied having the durable preference.
Does a verified plan mean a booking was purchased?
No. The source separates evidence states and leaves commercial transactions as future work.
More details
About this account
The owners supplied the YUL/YVR memory conflict and broken provider handoffs as observable failures. Codex repaired memory contracts and checked provider evidence, freshness and route-specific handoffs.
Firsthand dated engineering account opened. Codex build and Qwen runtime roles explicit. Published release/testing outcomes are creator-reported; planning evidence is not represented as a purchased or completed trip. Suggested prompt written by Agents In Use; no original task prompt was found in the verified text.
Credit: Mohamed Zidan and osama zaidan — Devpost
Source published
Added to Agents In Use
Source checked
Tools and topics
Codex CLI · Git worktrees · Playwright · PostgreSQL · Qwen Cloud
travel memory · booking handoff · evidence gates
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.