Spec-Kit + Copilot: large code output, weak constitution adherence, and “TODO = done” tasks

[bibin-aplxs]

Hi everyone,

I’m trying to understand whether I’m misusing Spec-Kit with GitHub Copilot, or if I’m hitting current limitations of the workflow.

Setup

Spec-Kit + GitHub Copilot

Model: Claude Opus 4.6

Goal: Monitoring dashboard app

Followed the docs strictly:

Created a detailed constitution (~1000 lines)

Used /speckit.specify

Iterated through plan → tasks → analyze

Started with Auth + Admin screens

Issues I’m facing

Huge code output: First feature alone generated ~10,000 lines for very basic functionality.

Constitution not fully followed: Explicit architectural and coding constraints were ignored.

Low-quality task completion: Many tasks were marked “done” with only // TODO comments.

Iterations didn’t converge: Multiple plan/task/analyze cycles still surfaced critical issues, and fixes caused regressions.

Questions

Are large constitutions (1000+ lines) a bad practice?

How do you prevent “TODO = completed task” behavior?

Is Spec-Kit mainly suited for small, incremental features?

Does model choice significantly affect constitution adherence?

Any real-world best practices to avoid code explosion and enforce constraints?

I like the spec-first idea, but so far the output feels high-ceremony with low signal.
Would love to hear from others who’ve made this workflow successful.

Thanks!