dev/PLAN.md

Next-Step Plan (Post Workstreams B → C)

Current State

• code-rs/ owns every forked crate; no crate depends on ../codex-rs (guarded by scripts/check-codex-path-deps.sh).
• codex-rs/ is synced to openai/codex@c3951e50 and treated as a pure upstream mirror.
• Diff tooling (scripts/upstream-merge/diff-crates.sh, highlight-critical-changes) highlights where our fork diverges.

Workstream C Priorities

1. Triage Critical Diffs

   • Review .github/auto/upstream-diffs/critical-changes/*.md.
   • For each crate (core, app-server, apply-patch, etc.), decide whether to adopt upstream wholesale, selectively port fixes, or keep the forked version.

2. Identify Re-export Candidates

   • Crates now identical between codex-rs and code-rs (e.g., git-apply, linux-sandbox, responses-api-proxy) can potentially be removed from code-rs in favor of upstream to reduce maintenance.
   • Document which crates can be dropped vs. which require fork-only patches.

3. Plan Merge Windows

   • Use the updated prompts/MERGE.md playbook for each upstream intake.
   • Prefer small, focused merges (e.g., TUI-only, CLI-only) followed by testing and guard verification.

4. Port Upstream Improvements

   • Where we lag (e.g., new config loader, execpolicy2, feedback view), cherry-pick or manually port upstream commits into code-rs so future merges shrink.
   • Keep fork-specific UX (themes, browser) while importing upstream bug fixes.

5. Clean Up Unused Crates

   • Large upstream-only crates (async-utils, app-server-test-client, windows-sandbox-rs, etc.) exist solely under codex-rs. Confirm they’re unnecessary for the fork and, if so, exclude them from future workstreams.

6. Automation

   • scripts/check-codex-path-deps.sh now runs automatically at the start of ./build-fast.sh, so both local builds and GH workflows that invoke it enforce the guard.
   • Continue regenerating diff/critical reports after every upstream merge to maintain visibility.

Success Criteria

• Regular merges from openai/codex:main proceed with minimal conflict.
• Critical upstream fixes/features are either upstreamed back or intentionally tracked in code-rs diffs.
• codex-rs/ remains a byte-for-byte mirror at every merge point.