Is Agents.md retarding creative use of AI in development?

Development
1

Is AGENTS.md too heavy of a burden for brainstorming a feature?

Prompted ChatGPT with:

I suspect that the Gramps AGENTS.md is being applied too early in the brainstorming a feature process. And that the extra burden is crippling what can be accomplished: `When I first started using Claude to wishcode a feature for Gramps, it allows lots of iterative revisions while building some fairly complex stuff. Now I consider myself luck if 3 evolutions as allowed before running out of tokens.

It makes me wonder if the AGENTS.md is not too much for early stages of development? It is as though each step has to be polished as though it was ready for public ready. That’s excessive AI labor for proof-of-concept stages.` Please comment on this theory.

ChatGPT’s Short answer

Yes — your theory is plausible. A long, “production-readiness” AGENTS.md can absolutely drag down early-stage AI-assisted feature exploration if it is injected into every turn, treated as always-current priority context, or interpreted as requiring polished, merge-ready code at every intermediate step.

The Gramps file is not unreasonable as a pre-merge contribution policy, but it is heavy as a brainstorming / proof-of-concept policy. GitHub shows the current AGENTS.md as 313 lines / 11.4 KB, and it includes commands, architecture, CI, formatting, type hints, docstrings, import grouping, class headers, logging, i18n, testing, commit-message rules, translation-file rules, error handling, branch rules, and PR rules. That is a lot of policy to carry while merely exploring an idea. (github.com)

Why the theory makes sense

  1. AGENTS.md likely consumes context before the feature idea even starts
  2. The rules mix three different phases of work
  3. The document encourages “final form” behavior too early

see the full response: