Codex CLI drops inline pasted payload when generating follow-up doc
What version of Codex is running?
codex-cli 0.47.0
What subscription do you have?
Pro
Which model were you using?
gpt-5
What platform is your computer?
Linux 6.6.87.2-microsoft-standard-WSL2 x86_64 x86_64
What issue are you seeing?
Observed behavior: during a documentation task, I supplied a multi-hundred character payload via inline paste. The assistant acknowledged receiving only a placeholder marker (e.g., “Pasted Content #### chars”) and proceeded without the actual content, leading to incomplete deliverables. Expected behavior: the assistant should ingest the pasted payload verbatim and make it available for downstream reasoning. Repro steps: paste a large text block in-line during a Codex CLI session, then ask the assistant to reference that content; observe that the assistant can only see the placeholder label instead of the real data. Impact: prevents accurate execution of tasks that rely on user-provided context. Request prioritization for restoring full payload access or clarifying limits surfaced to the user.
What steps can reproduce the bug?
- Open a Codex CLI session and begin a conversation with the assistant.
- Paste a multi-paragraph or long-form text block directly into the chat input (no file upload, just inline paste).
- Possibly take other steps editing before and after the pasted text placeholder.
- Submit the message and immediately ask the assistant to summarize or reference the pasted material.
- Observe that, instead of the actual text, the assistant only acknowledges a placeholder such as “Pasted Content #### chars,” demonstrating that the full payload was not ingested.
What is the expected behavior?
When a user pastes multi-line or large-form text directly into the Codex CLI conversation, the assistant should receive the full payload exactly as supplied—no truncation, placeholders, or metadata substitutions. The pasted content must be available for immediate reasoning, citation, and reuse in follow-up responses, so the assistant can produce accurate outputs that reflect the user’s source material.
Additional information
_No response_
3 Comments
Steps to repro (updated)
Thanks for reporting the issue.
I tried to repro the issue as described, and it appears to be working for me. I wonder if perhaps there's something specific about the contents you pasted. For example, does the text include uncommon Unicode characters? I tried pasting some big pieces of text from some wikipedia pages, and I was then able to ask the agent about the text.
Yes, admittedly reproducing this is hard without a detailed audit log of my exact session where it occurred. I tried asking codex if it had a self report tool to OpenAI headquarters, but understandably that's a security challenge to implement.
It MIGHT have been a copy/paste from Windows Chrome Gmail -> pasted directly into terminal wsl running codex. Or perhaps it was a copy-paste from another wsl terminal (it's been adding chars to the paste in some cases in my environment). HTH.