[Beta] I'd like to test

[BalaSankarPazhani]

I am a solo developer with active side projects (greenfield or partial) and have been using a similar approach.

My current workflow uses a similar pattern — a planning AI (Claude on claude.ai, that I call Claudio) produces versioned markdown specs covering PRD, Architecture, Database, API, and Frontend concerns. A separate implementation AI (Claude Code CLI, that I refer to as Codio) reads those docs before touching any code and follows a strict divergence protocol. I have ~3,300 transactions in production on this system.

What I want to learn from the beta: whether your three-file structure (requirements / design / tasks) maps cleanly onto a more granular six-document setup, and whether your test suite methodology is portable to validating my existing spec format.

Specific things I'll be testing: greenfield feature specs for a two-user auth system, migration sequencing assertions, and whether the Universal Instruction Block materially changes Codio's behaviour compared to my current handoff prompts.

I am interested in this skill and would like to see how this improves my throughput.

Thanks,
Bala

[FredAntB]

Hi Bala — First of all thank you so much for taking your time to open this issue and apologies for the late response.
I'm grateful for your interest in this project and I believe that your feedback on the three-file vs
six-document mapping is going to be genuinely useful.

A few things to orient you just in case:

Installation
Download the .skill file from the v1.0-beta release and install
via Claude Settings → Skills, or clone the repo and open it in
the Claude Code tab — CLAUDE.md bootstraps the skill automatically.

Your specific test cases
All three are well-suited to the skill:

• Greenfield two-user auth spec: use the new project interview
  flow. The skill will generate requirements.md, design.md, and
  tasks.md — compare how the three-file output maps against your
  six-document structure. Any gaps you find between the two
  approaches are high-value feedback.

• Migration sequencing assertions: tasks.md is where this lives.
  Each task gets an inline REQ reference and a Verify step —
  curious whether that structure handles migration ordering
  naturally or needs extension.

• Universal Instruction Block vs your current handoff prompts:
  this is the comparison we most want data on. Run both against
  Codio on the same task and note any behavioural differences.
  The UIB is in CLAUDE.md and .cursorrules — you can drop it
  alongside your existing docs without replacing them.

Filing feedback
Use the issue templates (beta-bug, beta-miss, beta-friction).
The single most valuable piece of data in any issue is the exact
phrase you used — especially for any case where the skill
doesn't trigger or produces unexpected output.

One ask: when you test the UIB against your current handoff
prompts, note which specific constraints Codio follows differently
(if any). That comparison is something no controlled test can
produce.

Welcome aboard — looking forward to your findings.