Skip to content

fix: say the answering model is unknown when ChatGPT does not name it - #112

Merged
youdie006 merged 1 commit into
mainfrom
fix/answer-model-from-conversation
Nov 18, 2026
Merged

youdie006 merged 1 commit into
mainfrom
fix/answer-model-from-conversation

Conversation

@youdie006

Copy link
Copy Markdown
Owner

Problem

Consult answers came back with model_used and pro_verified missing. A session that asked for Pro could not prove the answer came from Pro, and the response gave no reason, so "not checked" and "cannot be checked" looked the same.

Cause, measured 2026-11-17 on the live WSL container tab (read-only):

  • ChatGPT no longer renders data-message-model-slug. The page had none, and no other attribute matched model or slug.
  • The answer's action buttons (Copy, Rate response, Read aloud, Share, More actions) do not name a model.
  • The More actions menu lists only the time, Regenerate response, View sources and Branch conversation.

prodex reads rendered page content only (README "What it will not do", SECURITY.md, tests/architecture-boundaries.test.ts), so the answering model cannot be read. This change does not reverse that.

Change

  • model_used is "unknown" when the page does not name the model. This applies to the CLI JSON, the MCP pro_consult response, the consult_answer_saved receipt and the model_used: line.
  • A model_unverified warning says why pro_verified is left out.
  • The new picker_effort field records the step the picker's effort slider read once prodex had moved it, before the prompt was typed. The slider walk already confirmed this reading and then discarded it. It shows what was selected, not what answered, and the warning says so.
  • When ChatGPT names the model again, model_used and pro_verified are filled from it as before.
  • docs/clients.md and the MCP allow_model_fallback description name the new field and value.

Verified

  • New tests were watched failing first, then passing:
    • tests/chatgpt-browser.test.ts "returns the step the power slider read once the walk reached it"
    • tests/mcp-consult.test.ts "says the answering model is unknown when the page does not name it, and shows what the picker read"
    • "keeps quiet about the model when the page names it" (guard; it already passed)
  • Two existing expectations in tests/cli-pro-browser-send.test.ts now include the new warning for untagged answers.
  • npx tsc --noEmit clean. npm run release:verify (npm test, typecheck, build, package smoke, doctor) passed: release_verification=ok. The first full run, before the two expectations were updated, failed exactly those two tests (2006 passed, 2 failed).

ChatGPT stopped rendering data-message-model-slug (measured 2026-11-17 on the
live container tab; neither the answer's buttons nor its More actions menu name
a model), so model_used and pro_verified were silently missing and a Pro
consult could not be told from an unchecked one. prodex reads rendered page
content only, so the model cannot be read.

model_used is now "unknown" in that case, a model_unverified warning says why
pro_verified is left out, and picker_effort records the step the picker's
effort slider read once prodex had set it, before the prompt was typed. The
slider walk already confirmed that reading and discarded it.

Claude-Session: https://claude.ai/code/session_01EMqNwnnhHXznd2P2PDsaDV
@github-actions

Copy link
Copy Markdown

What this touches

  • browser adapter
  • CLI
  • MCP
  • tests

8 file(s), 143 line(s) changed. Tests changed.

This is an automated first pass, not a review. CI is the gate; a person decides.

@chatgpt-codex-connector

Copy link
Copy Markdown
Codex Review Summary: ✅ Completed

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review ✅ Completed 2026-11-18T03:25:44.017208Z c0bbd5a PR opened
ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

@youdie006
youdie006 merged commit d294d3e into main Nov 18, 2026
7 checks passed
youdie006 added a commit that referenced this pull request Nov 18, 2026
fix: port main PR #112 (say the answering model is unknown when ChatGPT does not name it) to the container image
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant