New chat creation fails with Invalid request: AbsolutePathBuf deserialized without a base path (existing tasks work)

Open 💬 2 comments Opened May 17, 2026 by MarcosiDCPrototypes

Environment

OS: Windows 11
VS Code Codex extension: openai.chatgpt-26.513.21555-win32-x64 (latest as of 2026-05-18)
Codex CLI: @openai/codex installed globally via npm, also up to date
Account: ChatGPT-authenticated (Business plan)

What happens
Opening the Codex panel and trying to start a new chat immediately shows a red banner: Error starting chat — Invalid request: AbsolutePathBuf deserialized without a base path. Continuing an existing task from the task list works correctly and returns proper GPT-5.5 responses, so auth/network/model routing are fine. Only the new-chat creation code path fails.
What we already ruled out

No [agents.*] blocks in ~/.codex/config.toml (so not #19257 / #19261)
No ~/.codex/agents/ folder
No project-level .codex/config.toml
Not the WSL agent mode (running Windows-native, sandbox = "unelevated") — so not #16815
Stripping \\?\ Windows long-path prefixes from [marketplaces.*] source lines does not help; restored after testing.
Reload Window can transiently clear the symptom but it recurs on next chat creation.

Notes
Symptom is similar to #19257 and #16815 but neither root cause applies. Looks like another AbsolutePathBuf deserialization site missing an AbsolutePathBufGuard, this time on the new-chat creation path rather than agent role config.
Config (sanitised)
tomlmodel = "gpt-5.5"
model_reasoning_effort = "high"
[windows]
sandbox = "unelevated"

8x [projects.*] blocks (all absolute paths, mixed-case drive letter)

[marketplaces.openai-bundled]
source = '\\?\C:\Users\<user>\.codex\.tmp\bundled-marketplaces\openai-bundled'
[marketplaces.openai-primary-runtime]
source = '\\?\C:\Users\<user>\.cache\codex-runtimes\codex-primary-runtime\plugins\openai-primary-runtime'

plugins enabled: browser-use, documents, spreadsheets, presentations

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗