Hide GPT-5.6 reasoning-summary empty placeholders

[pecigonzalo]

Problem

With openai-codex/gpt-5.6-terra and gpt-5.6-sol, visible thinking blocks sometimes show:

**Planning configuration reading**

<\!-- -->

The comment is present in the completed OpenAI reasoning item's summary, so Pi is not dropping readable reasoning. A comparison run with gpt-5.5 did not produce it.

Expected behavior

Do not render a summary part that is only a bold status heading followed by an empty HTML comment. Preserve literal comments in real content, for example Use `<\!-- -->` in JSX..

Prior art

OpenAI Codex handles this exact pattern as a standalone empty placeholder:

• https://github.com/openai/codex/blob/main/codex-rs/tui/src/history_cell/messages.rs#L501-L558
• https://github.com/openai/codex/blob/main/codex-rs/tui/src/history_cell/tests.rs#L2388-L2403

Suggested fix

Keep reasoning-summary parts separate until response.reasoning_summary_part.done. Suppress only the exact placeholder form from displayed thinking, while retaining the original reasoning item in thinkingSignature for replay.

Environment: Pi 0.80.6, openai-codex, hideThinkingBlock: false

[github-actions]

This issue was auto-closed. All issues from new contributors are auto-closed by default.

Maintainers review auto-closed issues daily and reopen worthwhile ones. Issues that do not meet the quality bar in CONTRIBUTING.md will not be reopened or receive a reply.

If a maintainer replies lgtmi on one of your issues, your future issues will stay open. If a maintainer replies lgtm, your future issues and PRs will stay open.

See CONTRIBUTING.md.

[badlogic]

need to fix that in the assistantmessagecomponent, can't filter out on the assistant message as that may break signature

[zqcli]

need to fix that in the assistantmessagecomponent, can't filter out on the assistant message as that may break signature

Hi

I looked into this and have a fix ready. Summary of the approach, would a PR for this be welcome ?

The placeholder parts (**heading** followed by a body that is only <!-- -->) are dropped at display time in AssistantMessageComponent, not in the message data. The original reasoning item is left untouched in
thinking/thinkingSignature, so provider replay is unaffected — this follows the note that filtering on the assistant message would break the signature.

Details:

• Added a pure helper stripEmptyReasoningPlaceholders(thinking) that splits the summary into parts and drops any part whose body is exactly <!-- -->, keeping the rest as-is.
• Keys off the content pattern, not the model id, so literal comments in real content are preserved (e.g. Use `<!-- -->` in JSX.). Comment-free reasoning returns unchanged via a fast path.
• Applied in both the visibility check and the render path so an all-placeholder summary collapses cleanly with no stray heading or blank line.
• Regression test under test/suite/regressions/6524-*.test.ts covering: placeholder dropped but real content kept, all-placeholder collapses to empty, literal <!-- --> in real content preserved. npm run check passes.

I kept it in the rendering layer rather than the provider stream or a compat flag, since this is display-only and compat flags are scoped to request shaping.

Thanks for your time.