Stripping out the harness and locking the workflow into documents

Publicly correcting the conclusion given four months earlier, of using different tools for different purposes, the write-up now says all plugins that enforce procedures have been removed and only one setting, thinking effort, gets adjusted. The reason for leaving just one line for the model and one for effort in the settings file is to keep visible what the model is doing at any moment, with the point that wrapping it in a framework means losing track of what that framework does. Instead, three-line decision records (decision, reason, rejected alternatives) are kept as one file per decision and read first at the start of each new task, and context files are placed separately in each subfolder rather than at the repository root. Repetitive work runs on 28 skills, Markdown documents that can be edited directly, and the fact that adding one line where something went wrong stops it from recurring is contrasted with frameworks built by others. However, the judgment of which model is better is itself one person's impression, so it is hard to take at face value.

This is a note: my own write-up after reading several sources. It has no source link; events checked against a source are kept as records.