Duplicate text output on codex

[theredsix]

• I searched existing issues and this is not a duplicate

Environment

Field	Value
Superpowers version	v5.0.7-7-g917e5f5
Harness (Claude Code, Cursor, etc.)	codex
Harness version	v0.120.0
Model	gpt-5.4 xhigh
OS + shell	linux, bash

Is this a Superpowers issue or a platform issue?

• I confirmed this issue does not occur without Superpowers installed

What happened?

Superpowers brainstorming duplicate text blocks during the "Ask user questions" phase. This happens about 25% of the time. This does not appear on the claude code version of this plugin.

Steps to reproduce

1. Install superpowers on codex
2. Use the brainstorming plugin on a code base with an ambiguous question
3. Observe the text duplication issue

Expected behavior

Text blocks should only appear once.

Actual behavior

Text block appears multiple times.

Debug log or conversation transcript

[obra]

Huh. I'm not seeing this in Codex. Does it happen for you without Superpowers installed?

[Diiiiooo]

I’ve also encountered this bug, but I’m not sure if it’s related to “superpowers”.

[theredsix]

@obra I haven't encountered it without Superpowers, it appears to only happen during superpowers plan and at the frequency of 25% of the plans.

[ronald-weller-hogrefe]

I can second this. On Windows, seems only to happen with Superpowers but inconsitently.

[signingoff]

same here, not sure why.
Windows + Codex

[TheQjc]

I've encountered this before, using the CodeX CLI on Windows.

[vesselin-ag]

I'm also experiencing the same bug. I just installed Superpowers fresh in Codex CLI (v0.120.0) on macOS (26.4.1) and the first time I used the brainstorming skill it started duplicating questions

[Sisyphean-a]

Confirming this bug on my end. I'm seeing duplicate text blocks across the commentary and final sections. It seems that after the initial occurrence, the system gets stuck in this state, as the duplication rate for subsequent responses jumps to nearly 100%. This makes the tool almost unusable once the pattern starts.

[vesselin-ag]

For now, I added this to my AGENTS.md and so far I haven't seen duplicate output:

## Interaction Rules

- Ask a clarifying question only once.
- When work is still in progress, use a single user-visible progress or question message. Do not repeat the same content in both an intermediate update and a completion message.
- Do not mirror a `commentary` question in `final`.
- Use `final` only when the current request is actually complete, or when you are intentionally handing control back with a concise status that does not duplicate a prior question.
- During brainstorming, debugging, or planning workflows, each step should produce one question at a time and then wait for the user's reply.

## Working Rules

- Prefer repo-local instructions in this file over generic defaults when they conflict.
- Keep progress updates short and non-redundant.
- If a workflow instruction appears to require behavior that would duplicate visible output, prefer the non-duplicating interpretation.