tdd-guard backed coding with Sonnet (or event opus) used in the change validation/check burns a lot of tokens and as Anthropic seems to regulary cut the limits a bit more (at least it feels like it), I experimented with using GLM 4.7 (Z.ai) as a drop-in-replacement. GLM models are much more cheaper to use (magnitudes cheaper) and this wont affect the actual code quality.
The instructions.md file requires heavy changes as the GLM 4.7 model requires vastly differenct prompting techniques. I made some iterations and found a prompt that works quite well. (200 lines vs 50 lines ... shows how "good" GLM will be for the "normal" coding)
I suggest to make an addition to the documentation with a new "section" (documentation file) in this repo, briefly explaining how to use GLM with the .env file (specific how to do in tdd-guard) and with the adapted instructions file content. If that sounds reasonable, I will create a PR.
tdd-guard backed coding with Sonnet (or event opus) used in the change validation/check burns a lot of tokens and as Anthropic seems to regulary cut the limits a bit more (at least it feels like it), I experimented with using GLM 4.7 (Z.ai) as a drop-in-replacement. GLM models are much more cheaper to use (magnitudes cheaper) and this wont affect the actual code quality.
The
instructions.mdfile requires heavy changes as the GLM 4.7 model requires vastly differenct prompting techniques. I made some iterations and found a prompt that works quite well. (200 lines vs 50 lines ... shows how "good" GLM will be for the "normal" coding)I suggest to make an addition to the documentation with a new "section" (documentation file) in this repo, briefly explaining how to use GLM with the .env file (specific how to do in tdd-guard) and with the adapted instructions file content. If that sounds reasonable, I will create a PR.