Codex / agent CLI task

# DevAgentRadar → OpenAI Codex

Task: evaluate and optionally implement follow-ups from this coding-assistant release.

## Context
Assistant: OpenAI Codex CLI
Proposal: Adopt Codex `/export`, `codex exec fork`, and thread cost status
Summary: Codex CLI rust-v0.148.0 adds Markdown export via `/export`, session fork/archive/restore, startup prompt drafting, thread credit/cost in `/status`, Bedrock Runtime, and async/MCP hooks—use them to make session reuse, cost visibility, and automation first-class.
Primary source: https://github.com/openai/codex/releases/tag/rust-v0.148.0

## Why it matters
This release turns ephemeral TUI chats into durable artifacts and branchable work. `/export` writes full conversations to Markdown (clipboard or file), so reviews and handoffs no longer depend on scrollback. `codex exec fork` plus archive/restore in the resume picker lets you branch an approach without losing the parent thread, and resumed sessions now restore cwd and approval policy more reliably. Estimated credits/cost in `/status`, status lines, and terminal titles give eligible workspaces live spend signals before a long run. Bedrock Runtime support (AWS profile, region, GPT-5.6 routing) matters if you standardize on AWS. Hooks can run async commands and call MCP tools, so post-turn automation can leave the critical path. Pair these with the sandbox fail-closed path fixes and model-switch instruction cleanup so forked/exported workflows stay predictable.

## Suggested actions
1. Upgrade Codex CLI to rust-v0.148.0, open a real task in the TUI, then run `/export` and save the full transcript to `./docs/codex-sessions/$(date +%Y%m%d)-<topic>.md` (also try clipboard export once to confirm both paths).
2. From a finished useful thread, run `codex exec fork` to branch an alternate implementation; in the TUI resume picker, archive the parent session and later restore it to verify cwd, approval policy, and transcript preview come back intact.
3. During the next multi-step run, watch terminal title and status line for estimated thread credits/cost, then run `/status` mid-turn and before exit; if spend is higher than expected, stop and fork a narrower follow-up instead of continuing the same thread.
4. If your team uses AWS, configure Amazon Bedrock Runtime as the built-in provider with your AWS profile and region, route a non-prod GPT-5.6 smoke prompt through it, and compare `/status` cost/credits against your default provider on the same prompt.
5. Add or update a Codex hook so a post-turn step runs asynchronously and invokes an MCP tool (for example: format export Markdown or open a tracking issue) without blocking the TUI; confirm the main turn finishes while the hook continues.

## Config surfaces this release may change
- skills **(high confidence)** — check your repo before applying
- sandbox settings — check your repo before applying
- MCP servers — check your repo before applying


Deliverables:
- Short impact assessment
- Optional patch plan (files + steps)
- Do not invent APIs not in the source notes