Work Log - 2026-06-16

🎯 Focus for Today¶

Local mock/dev env for working on oc-opsdevnz module
Focusing now on some dev and release proceses
Setup OIDC based release process for oc-opsdevnz

✅ What Got Done¶

Updated module-template-opsdevnz: switched CI to uv, added build dep, added py.typed marker (PEP 561). See PR #3.
Published oc-opsdevnz v0.2.5 to Test PyPI and PyPI via OIDC.

🧠 Notes & Reflections¶

Setting up a mock development environment for the oc-opsdevnz module in this repo, we treat the OpenCollective staging env as a UAT env, so need another test environment to do actual work. Obviously. Immediately discovering some weirdness is this very rough early dev version of our tool. You could ask to check what hosts you have setup, and point it at your collectives configuration and it would happily report to you the hosts. Ultimately this is just some early fail validation checks and doesn't hurt to have them. Made some tests more generic, and now we just want to smooth out the release process so a small patch version update for oc-opsdevnz today.

Also gave Kimi K2.7 a spin. One of the things I like to use the LLMs for is for creating a good descriptive pull request, and Kimi struggled a bit to get this right. From a tools perspective, nothing about writing the actual PR itself, it of course does a good job on that, but tried to use bash at first, and then ended up putting a payload in /tmp with the description. Seems like a tools gap we need to fix.

I had DeepSeek review Kimi's work and it approved. Mentioned we should look at adding some testing to the --only flag next.

I stumbled a bit over the pipeline setup, just forgot to add --extra dev in the publish.yml TODO: Need to go back and check how that is set in the template file.

Different ways you can set that up, but I feel like build deps are dev things.

I asked DeepSeek if anything else should be added in the build process and it mentioned adding the py.typed marker to help downstream consumers of the module check types, since annotations are used per PEP484, and read a discussion on GitHub about how it's at least harmless.

⏳ Mañana¶

Standardize AI assistant PR/MR description workflow (gh/glab templates)

🔥 Token Burn¶

Kimi K2.7 Code

250k tokens
Rolled over the context window
$5.65 spent

DeepSeek V4 Pro code review and follow ups

65k tokens
7% context used
$0.24 spent