Skip to content

chore(upstream): accept Codex 0.157.1 - #58

Draft
Juliusolsson05 wants to merge 1 commit into
mainfrom
chore/accept-codex-0157
Draft

Juliusolsson05 wants to merge 1 commit into
mainfrom
chore/accept-codex-0157

Conversation

@Juliusolsson05

@Juliusolsson05 Juliusolsson05 commented Sep 27, 2026 •

Copy link
Copy Markdown
Owner

Fixes #16 (accepting 0.157.1 with its known gaps filed; see below). App side: Juliusolsson05/agent-code#234.

Compatibility review for 0.149.1 → 0.157.1. The full evidence is in agent-code docs/plans/2026-09-26-accept-codex-0157.md (branch chore/accept-codex-0157).

Checklist

  • Release notes reviewed (19 releases). The relevant items:
    • #47096 moves some TUI hints above the composer. The idle ? for shortcuts hint row this package reads is unchanged in the 0.157.1 recording.
    • #47113 and #41912 add rollout content (see the corpus survey below).
    • Rollout compression (0.152/0.153): none of the 1,850 local files is compressed.
  • Runtime starts and attaches: Agent Code has driven 0.157.1 panes daily since 2026-09-25, and the proxy and composer work in feat(composer): live, attribute-aware composer state (agent-code#800, #1313) #54/Settled plain screen for text proofs of an empty composer #55/fix(proxy): bound the events mirror and take it off the synchronous path #56/fix(composer): a draft taller than 12 rows is still a draft #57 was recorded against 0.157.x.
  • Transcript and session paths:
    • A type-by-version survey of 1,850 local rollouts (208 of them 0.157.x) found one new top-level type, token_usage_record (from 0.153), which is filed as agent-transcript-parser#38.
    • 0.157 writes no event_msg:user_message. The prompt is a role-user response_item plus item_completed/UserMessage. FreshRolloutClaim matches every durable user observation, response_item included, so fresh ownership still finds it, and SessionList reads role-user response items.
  • Screen parsers: raw 0.157.0 and 0.157.1 PTY recordings pin the composer classifier (idle, a draft, Ctrl+C, and a 20-line draft, which uses the new fullscreen default). The trust and approval parsers have no 0.157 recording.
  • Fixtures and suites: CI on this PR.
  • Rollout shape change filed: agent-transcript-parser#38.

Known gaps found in review

Each is filed. All but #1289 are confirmed inherited from the already-accepted 0.149.1, so accepting 0.157.1 does not make them worse:

  • codex-headless#59: listCodexSessions summaries show the injected AGENTS/environment block. Agent Code's picker reads Codex's own index and is unaffected.
  • codex-headless#62: the semantic tool lifecycle ignores item_completed (CommandExecution/McpToolCall), the only tool form since 0.149.
  • agent-code#1362: read_agent_transcript drops custom_tool_call exec commands.
  • agent-code#1363: the no-index conversation fallback reads only event_msg:user_message.
  • agent-code#1289 (existing): 0.157 compacted payloads have no type, so no compaction boundary renders.
  • codex-headless#60: the recorded rollout-ownership tests run past 5 s under load.
  • agent-transcript-parser#38: token_usage_record is uncatalogued.

Unverified, stated as residual

0.157.0 #47178 made the fullscreen transcript the default. The composer classifier is pinned in that mode: the 0.157.1 recording uses the alternate screen. Trust and approval overlays have no 0.157 recording.

🤖 Generated with Claude Code

Compatibility review for #16 (agent-code#234): release notes 0.149.1 to
0.157.1, a type-by-version survey of 1,850 local rollouts (only new top-level
type: token_usage_record, filed as agent-transcript-parser#38; the prompt now
arrives as a role-user response_item plus item_completed, which fresh-rollout
ownership already matches), 0.157.0/0.157.1 PTY recordings pinning the
composer classifier, and daily 0.157.1 use in Agent Code.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@Juliusolsson05

Copy link
Copy Markdown
Owner Author

Blocked — do not merge. The acceptance review did its job: it found three real 0.157.x breaks, all binary-verified and none inherited from 0.149.1:

  • codex-headless#63: the prompt-input profile refuses every Codex after 0.149.1 (unsupported-cli at 0.157.1), so fresh-rollout ownership through prompt evidence is dead;
  • codex-headless#64: the write-stdin approval overlay (0.151+) is never detected, so approval automation hangs;
  • codex-headless#65: 0.156 reworded the trust dialog to "Trust this folder?", so every trust detector misses it on the fallback path.

The accepted version stays at 0.149.1 until those land. This PR is converted to draft and will be rebased onto the fixes. The earlier known-gaps list still stands.

@Juliusolsson05

Copy link
Copy Markdown
Owner Author

Parked (W1, 2026-09-27), not abandoned. This bump waits on the two 0.157.1 blockers the review found:

  • codex-headless#65, trust dialog: fixed in codex-headless#67. All three verifications are in, and one reviewer re-verification is running.
  • codex-headless#63, prompt-input profile: fixed in codex-headless#69, awaiting its three reviews.

codex-headless#64 (write-stdin approval) is no longer a blocker: the overlay is behind the off-by-default write_stdin_approval flag in 0.157.1 (evidence on #64).

Once #67 and #69 merge, this PR takes their pointers, re-runs the acceptance gate and leaves draft. New 0.157 findings from that work are tracked separately: #66 (background server) and #68 (a ? dropped from single-line prompts).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

maintenance Maintenance work provider:codex Affects Codex integration

Projects

None yet

Development

Successfully merging this pull request may close these issues.

chore(upstream): Codex 0.160.1 is newer than accepted 0.149.1

1 participant