# Tracking feedback with Claude Code

DevFeed: [Tracking feedback with Claude Code](<https://devfeed.tech/articles/tracking-feedback-with-claude-code-52451.md>)

Original publisher: [Read original article](<http://www.jackfranklin.co.uk/blog/claude-code-ai-feedback-skill/>)

Published: 2026-04-24T00:00:00Z

Content type: article

Language: en

Sources: [Jack Franklin](<https://devfeed.tech/sources/jack-franklin.md>)

Topics: [Claude Code](<https://devfeed.tech/topics/claude-code.md>), [pair\_programming](<https://devfeed.tech/topics/pair-programming.md>), [Agent Skills](<https://devfeed.tech/topics/agent-skills.md>), [Tool](<https://devfeed.tech/topics/tool.md>), [To-Do](<https://devfeed.tech/topics/todo.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [bug](<https://devfeed.tech/tags/bug.md>), [bugs](<https://devfeed.tech/tags/bugs.md>), [claude](<https://devfeed.tech/tags/claude.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [debugger](<https://devfeed.tech/tags/debugger.md>), [feedback](<https://devfeed.tech/tags/feedback.md>), [game](<https://devfeed.tech/tags/game.md>), [markdown](<https://devfeed.tech/tags/markdown.md>), [markdown-files](<https://devfeed.tech/tags/markdown-files.md>), [sqlite](<https://devfeed.tech/tags/sqlite.md>), [tokens](<https://devfeed.tech/tags/tokens.md>), [tracking](<https://devfeed.tech/tags/tracking.md>), [ui](<https://devfeed.tech/tags/ui.md>)

## AI overview

For playtesting OnTrack, the author first used Claude Code to collect notes in Markdown, where Claude grouped feedback, detected duplicates, and refined related items. As the lists grew to thousands of lines, even simple queries required reading whole files. The author replaced that workflow with a custom skill backed by a Deno CLI and SQLite: records hold titles, details, priority, status, category, project, and a completion flag. Completed items remain available as history but are hidden from default listings. The skill guides Claude to list titles and IDs before fetching details, check for related feedback before adding items, and infer the project from context when clear. The excerpt says this made targeted backlog queries faster and more accurate; it does not provide broader performance measurements.

## Source excerpt

I've been using Claude Code as a pair programmer, debugger, and feedback collector while I playtest my game, OnTrack. The setup has worked so well that I ended up building a small custom skill to make it scale and in this post I'll take you through how it works. (That last DevBlog was in 2024 (!), but I have continued to work on the game -- with AI my pace and progress has quickened, even if overall I've less time since becoming a dad.) Pausing development to playtest After a period of development, I like to do a playtest. I resist fixing things, and instead I play the game for an hour, noting down bugs, thoughts and suggestions as I go. I used to do this in a Trello board, but I recently decided to try using Claude as my bug collecting coworker and collate feedback in a markdown file. This worked really well! One of the benefits of using Claude is that it was able to detect similar feedback, or duplicates, or re-write old feedback to incorporate new points that were related. It also grouped them as it went under different categories such as UI, mechanics, or "quick wins". The problem of large Markdown files As I then followed these playtesting sessions with fixes, I asked Claude to mark feedback as complete by moving it into a separate document. In one hour of playtesting I would easily pick up on 50+ issues I wanted to fix (the game is far from done!) and one day I realised that the Markdown file with incomplete items was 3,000 lines long, and the one with completed items was 4,000 lines long! Claude was managing this fine - but I could see that as time grew, this list would become hard to manage, eat up more of my tokens, and become inefficient. Asking Claude to list the outstanding issues, or all the issues under a particular category, required it to read the entire Markdown file, even if I was only asking about a subset. The solution: SQLite I needed a way to store this data in a more structured form where Claude could query it more efficiently -- for example, ge