Ticket is surprisingly token-efficient compared to other CLI issue trackers

[a3lem]

Having burned out on Beads after the latest breaking change – canning sqlite+jsonl+daemon in favor of Dolt – I've resumed my search for an issue tracker that does 'just enough'.

Last night I compared how a bunch of options, including ticket, compare when put in the hands of Claude Code, particularly on the task of creating issues and defining dependencies. Very simple experiment setup. To the system prompt I appended the output of <command> --help and (if available) the output of <command> prime. The user message was a general instruction to create tasks based on the following text. The text itself was a paragraph about how to repot a plant. For the model I used Sonnet 4.6.

I repeated the experiment for each of these: beads, beads_rust, beans, dots, and ticket.

What I observed over and over is that, with Ticket, Claude Code uses up to 86% fewer tokens to achieve the same goal. In terms of total tokens used (system + messages), Dots beats Ticket by 100 tokens on average. When I repeated the experiment with Haiku, though, Dots tended to 'thrash about' with invalid CLI calls, whereas Ticket worked the same. (My theory is that the design of Ticket's CLI is slightly more intuitive, not just to me as a human but for the model also.)

tool	system prompt tokens	messages tokens
ticket	4.1k	1.7k
beans	5.4 k	4.6k
beads_rust	4.2k	7k
dots	3.7k	2k

So I'll be adopting Ticket!

Also want to note that this project feels the least 'sloppy' compared to alternatives.

1. Actual tests, which also introduced me to something new: behave + gherkin language.
2. The core vs extensions distinction
3. I can read the code and actually understand what it's doing, even if I'm no bash/awk expert. It gives me the confidence that I can fork it and modify/add anything I'm missing.
4. I 100% agree with the author's worse-is-better takes on twitter.