Skip to content

release: v0.51.53 — a run completion stops turning one turn into gigabytes - #1595

Merged
artokun merged 1 commit into
mainfrom
release/0.51.53
Aug 15, 2026
Merged

release: v0.51.53 — a run completion stops turning one turn into gigabytes#1595
artokun merged 1 commit into
mainfrom
release/0.51.53

Conversation

@artokun

@artokun artokun commented Aug 15, 2026

Copy link
Copy Markdown
Owner

Release cut carrying #1516 (PR #1521, contributed by @JusticeWay).

A Codex panel session reached ~3.02 GB and stopped replying: two foreign run completions at startup produced a 105,703,660-byte record with 28 inline images, and journal replay produced another one just like it.

Foreign/unidentified completions now arrive as text with their pixels detached — matching the correlation preamble, which already calls those runs UNDETERMINED. A matched completion attaches at most 8 images per turn, corrected at the drain. Withheld outputs are named with get_image coordinates; outputs with no filename say so and point at get_history.

Post-merge follow-up in this cut: the disclosure now also accounts for unnamed outputs on a MIXED event, where one named sibling had kept the generic "attached below" wording alive and made attached+omitted fail to sum to the total the same sentence quoted. Verified by mutation.

🤖 Generated with Claude Code

…bytes

A Codex panel session reached about 3.02 GB of private memory and stopped
producing a reply. On 0.51.29 a fresh session with zero compactions reached
211,885,110 bytes after two turns: two FOREIGN run completions at startup
produced one 105,703,660-byte record holding 28 inline images, and the
interrupted turn returned both events to the journal, so replay produced another
~105,701,700-byte record with the same 28 images (#1516).

The correlation preamble already declares a foreign or unidentified completion
UNDETERMINED and forbids treating it as the agent's awaited run. Attaching its
pixels contradicted that, so those completions still arrive as TEXT but no longer
put their images on an agent turn. A matched completion attaches at most 8 images
per TURN — a per-event cap was not a cap, because everything queued drains into
one turn — and the drain corrects the count rather than trusting it, since a
batch already in flight can be requeued behind a newer one.

Nothing is dropped silently. Withheld outputs are named with the coordinates
get_image needs, and where an output has no filename at all the text says so and
points at get_history instead, rather than offering a fetch that cannot be
addressed. Every output remains visible in the panel, and explicit user image
attachments and the journal's acknowledge/replay semantics are unchanged.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Copilot AI balanced review requested due to automatic review settings August 15, 2026 03:49

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

@artokun
artokun merged commit 09ab4f6 into main Aug 15, 2026
5 of 6 checks passed
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.

2 participants