Switch Codex CLI to GPT-6-Astra and isolated `--worktree` checkouts
Upgrade to Codex CLI rust-v0.154.0, pick GPT-6-Astra, run new/forked sessions in experimental--worktree//worktree checkouts, answer questions inline, share the Windows background server, and stop calling the removed codex mcp-server entry point.Why this loop
rust-v0.154.0 changes the model you should run and how parallel sessions should be isolated. GPT-6-Astra is in the model picker and Amazon Bedrock catalogs, and the bundled OpenAI Docs skill now includes Astra migration, compatibility, and prompting guidance—so prompts and Bedrock routing should move to Astra. Experimental worktrees create isolated checkouts for new or forked sessions via --worktree or /worktree, which you can browse and resume instead of colliding in one dirty tree. Inline questions let you pick a suggested choice or type custom text without losing the main draft, so answer in place rather than stopping the run. Windows sessions can share a background Codex server with daemon lifecycle commands and managed updates. Fresh sessions and forks follow server model defaults unless you override them; remote resume/fork keeps saved permissions. The deprecated codex mcp-server entry point is gone, so leftover MCP launchers will fail. MCP OAuth refresh is coordinated and surfaces login challenges instead of replaying rejected tool calls. Do not assume anything from the 54 omitted changes.
Proposed actions
- Upgrade OpenAI Codex CLI to rust-v0.154.0, select GPT-6-Astra in the model picker (also listed in Amazon Bedrock catalogs), and apply the bundled OpenAI Docs skill’s GPT-6-Astra migration, compatibility, and prompting guidance before rewriting prompts.
- For new or forked sessions, create isolated checkouts with
--worktreeor/worktree, then browse and resume those worktrees instead of sharing one working tree across parallel Codex runs. - When Codex asks a question while still working, answer inline with a suggested choice or custom text so the main draft is preserved and the run continues.
- Delete every script, IDE MCP config, and launcher that still invokes
codex mcp-server(removed in rust-v0.154.0). After upgrade, complete any MCP OAuth login challenge if token refresh fails, and do not expect rejected tool calls to replay automatically. - On Windows, attach sessions to the shared background Codex server and use daemon lifecycle plus managed-update commands instead of one server per session. On resume/fork, keep saved permissions and only override the model when you do not want the server default.
Agent prompt
Paste into your agent or query via MCP (get_agent_prompt) — free, no extra AI costCodex / agent CLI task
DevAgentRadar → OpenAI Codex
Task: evaluate and optionally implement follow-ups from this coding-assistant release.
Context
Assistant: OpenAI Codex CLI
Proposal: Switch Codex CLI to GPT-6-Astra and isolated --worktree checkouts
Summary: Upgrade to Codex CLI rust-v0.154.0, pick GPT-6-Astra, run new/forked sessions in experimental --worktree//worktree checkouts, answer questions inline, share the Windows background server, and stop calling the removed codex mcp-server entry point.
Primary source: https://github.com/openai/codex/releases/tag/rust-v0.154.0
Why it matters
rust-v0.154.0 changes the model you should run and how parallel sessions should be isolated. GPT-6-Astra is in the model picker and Amazon Bedrock catalogs, and the bundled OpenAI Docs skill now includes Astra migration, compatibility, and prompting guidance—so prompts and Bedrock routing should move to Astra. Experimental worktrees create isolated checkouts for new or forked sessions via --worktree or /worktree, which you can browse and resume instead of colliding in one dirty tree. Inline questions let you pick a suggested choice or type custom text without losing the main draft, so answer in place rather than stopping the run. Windows sessions can share a background Codex server with daemon lifecycle commands and managed updates. Fresh sessions and forks follow server model defaults unless you override them; remote resume/fork keeps saved permissions. The deprecated codex mcp-server entry point is gone, so leftover MCP launchers will fail. MCP OAuth refresh is coordinated and surfaces login challenges instead of replaying rejected tool calls. Do not assume anything from the 54 omitted changes.
Suggested actions
- Upgrade OpenAI Codex CLI to rust-v0.154.0, select GPT-6-Astra in the model picker (also listed in Amazon Bedrock catalogs), and apply the bundled OpenAI Docs skill’s GPT-6-Astra migration, compatibility, and prompting guidance before rewriting prompts.
- For new or forked sessions, create isolated checkouts with
--worktreeor/worktree, then browse and resume those worktrees instead of sharing one working tree across parallel Codex runs. - When Codex asks a question while still working, answer inline with a suggested choice or custom text so the main draft is preserved and the run continues.
- Delete every script, IDE MCP config, and launcher that still invokes
codex mcp-server(removed in rust-v0.154.0). After upgrade, complete any MCP OAuth login challenge if token refresh fails, and do not expect rejected tool calls to replay automatically. - On Windows, attach sessions to the shared background Codex server and use daemon lifecycle plus managed-update commands instead of one server per session. On resume/fork, keep saved permissions and only override the model when you do not want the server default.
Config surfaces this release may change
- MCP servers (high confidence) — check your repo before applying
After you finish
Do not report this as applied to DevAgentRadar. You cannot write the visitor's loop.
Tell the human: open https://devagentradar.com/proposals/openai-codex-rust-v0-154-0-switch-codex-cli-to-gpt-6-astra-and-isolated- and mark Applied, Skipped, or Failed. Proposal id: d1e2ba35-6a41-41c9-bbbb-e506003a8f00
Deliverables:
- Short impact assessment
- Optional patch plan (files + steps)
- Do not invent APIs not in the source notes
Your loop
This browser · no sign-in · not shared as “you”After you run the prompt
Only you can mark this. Agents cannot write your loop.
Your decision stays on this device. A public tally appears after a few votes.