Skip to content

docs: Codex teleport — design note (blocked on validation, no code)#3

Open
satelerd wants to merge 2 commits into
feat/teleportfrom
feat/codex
Open

docs: Codex teleport — design note (blocked on validation, no code)#3
satelerd wants to merge 2 commits into
feat/teleportfrom
feat/codex

Conversation

@satelerd

Copy link
Copy Markdown
Owner

Status: design note, NOT implemented

Codex teleport is blocked on validation and this PR adds no code — just docs/teleport-codex.md with the plan.

Why no implementation: building it needs a real Codex session to learn the on-disk format + resume semantics, and a Codex env to resume on. Right now there's no codex on the mini and the MacBook is off. Inferring the format and shipping code would repeat the bug PR #1 just fixed (Claude's project-dir encoding wasn't what anyone would guess; the assume-the-format path stayed green in tests but broke real resume).

What the doc covers: the grounding step (generate + inspect a real rollout), the open questions to confirm (where sessions live, the cwd field, how resume finds a dropped-in transcript), an implementation sketch reusing the existing repo-handling, and a non-circular test plan.

Next session

Start on a Codex-capable machine: run the grounding step, then implement + validate end-to-end (a real laptop→mini Codex teleport that resumes) — same bar Claude teleport met.

Base is feat/teleport (PR #1). Retargets to master after #1 merges.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Y1zMeQM7WsMbT5eK8eR8mH

satelerd and others added 2 commits June 23, 2026 00:49
Plan only — no code. Codex teleport needs the real session format + resume
semantics confirmed against an actual session before implementing; shipping
inferred-format code would repeat the Claude encoding bug. Blocked now: no codex
on the mini, MacBook powered off (no real session to package, no env to resume
on). Documents the grounding step + implementation sketch for the next session.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y1zMeQM7WsMbT5eK8eR8mH
Inspected real Codex rollouts on the MacBook. Confirmed: sessions at
~/.codex/sessions/YYYY/MM/DD/rollout-<ts>-<uuid>.jsonl; lines are
{timestamp,type,payload}; cwd is in the first line's payload.cwd (single place);
session keyed by UUID in the filename (no encoded-cwd dir → no path-encoding
trap). One open question for impl: how `codex resume` locates a session. Test
plan: codex+auth on the mini (copy auth.json), real e2e resume. Start here next.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Y1zMeQM7WsMbT5eK8eR8mH
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant