Migration Diary: Freeze a Golden Task Fixture Before You Leave a Paid Coding Agent

Description and use

Create a small, maintainable golden task fixture schema to capture the acceptance criteria previously defined by AI agent chat transcripts, and use it to validate quality after migrating to a new stack.

Problem

When a paid coding agent subscription ends, the team loses the chat‑based acceptance surface that defined what "done" looked like, risking quality drift on the new stack.

Outcome

The team now has a lightweight, versioned test suite that captures the agent’s implicit acceptance criteria, allowing them to safely cancel the paid subscription without risking quality regression.

Workflow

• 1. Identify recent agent‑drafted changes from merge history and write concise task descriptions.
• 2. Freeze pre‑change files as inputs next to each task prompt.
• 3. Create a runner that applies the frozen inputs, invokes the new local agent command, and asserts file changes and test results.
• 4. Execute the runner to ensure tasks pass; failures indicate destination drift.
• 5. Commit the fixture suite and runner to the repository for ongoing regression testing.

Evidence

A frozen fixture turns those unspoken expectations into files you can rerun on cheaper local or remote infrastructure.

Source

The paid agent also absorbed style rules, review nits, and file‑ban instincts that never landed in git.

Source