Copilot/Spec-kit seem unable to write PowerShell scripts that work.

[BurtHarris]

Config: Copilot in vscode running GPT-5 Lite. It seems more reliable to write Python.

I'm using the PowerShell version of the copilot scripts updated yesterday. It gets involved in loops trying to correct PowerShell errors, frequently due to improper use of commas, when I give it a prompt as simple as commit. The darn thing keeps trying to use PowerShell for tasks it's not suited for. I finally restricted its access to the terminal tool and told it to use the other tools available to it.

I have tried to address this by tuning the constitution and copilot_instructions.md.

[BurtHarris]

Why can't I attach the markdown documents?

Failed to upload "constitution.md"<

P.S. I'm a retired Microsoft Dev, shipped Microsoft English Query back in SQL 2000. The NLP was implemented in a compiled Lisp-like language, and hand-tuned linguistic knowledge coded in Lisp, aided by a semantic model that combined a database and language applicable to the customer's domain.

MEQ knew when it didn't know how to answer a question. My contribution was a tool that uses a set of Lisp-generated 'options' on how to interpret unrecognized terms, and maps them into the database without requiring an expert; someone who knew the DB schema could generally help it out. The classic question back then, from testers, was 'Who bought chai' against the Northwind Traders database. A DBA walked through a set of questions and, for example, mapped 'Who' to the customer's table, 'What' to the transaction table, and 'Chai' to an instance out of the products table. This extended the semantic model, enabling Lisp to answer similar questions moving forward.

[github-actions]

This issue has been automatically marked as stale because it has not had any activity for 150 days. It will be closed in 30 days if no further activity occurs.